plugins/gpu_neon/psx_gpu/psx_gpu_simd.c

   1 /*
   2  * Copyright (C) 2011 Gilead Kutnick "Exophase" <exophase@gmail.com>
   3  * Copyright (C) 2022 Gražvydas Ignotas "notaz" <notasas@gmail.com>
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation; either version 2 of
   8  * the License, or (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  */
  15
  16 #include <string.h>
  17 #include "psx_gpu.h"
  18 #include "psx_gpu_simd.h"
  19 //#define ASM_PROTOTYPES
  20 //#include "psx_gpu_simd.h"
  21 #ifdef __SSE2__
  22 #include <x86intrin.h>
  23 #endif
  24 #ifndef SIMD_BUILD
  25 #error "please define SIMD_BUILD if you want this gpu_neon C simd implementation"
  26 #endif
  27
  28 typedef u8  gvu8   __attribute__((vector_size(16)));
  29 typedef u16 gvu16  __attribute__((vector_size(16)));
  30 typedef u32 gvu32  __attribute__((vector_size(16)));
  31 typedef u64 gvu64  __attribute__((vector_size(16)));
  32 typedef s8  gvs8   __attribute__((vector_size(16)));
  33 typedef s16 gvs16  __attribute__((vector_size(16)));
  34 typedef s32 gvs32  __attribute__((vector_size(16)));
  35 typedef s64 gvs64  __attribute__((vector_size(16)));
  36
  37 typedef u8  gvhu8  __attribute__((vector_size(8)));
  38 typedef u16 gvhu16 __attribute__((vector_size(8)));
  39 typedef u32 gvhu32 __attribute__((vector_size(8)));
  40 typedef u64 gvhu64 __attribute__((vector_size(8)));
  41 typedef s8  gvhs8  __attribute__((vector_size(8)));
  42 typedef s16 gvhs16 __attribute__((vector_size(8)));
  43 typedef s32 gvhs32 __attribute__((vector_size(8)));
  44 typedef s64 gvhs64 __attribute__((vector_size(8)));
  45
  46 typedef union
  47 {
  48   gvu8  u8;
  49   gvu16 u16;
  50   gvu32 u32;
  51   gvu64 u64;
  52   gvs8  s8;
  53   gvs16 s16;
  54   gvs32 s32;
  55   gvs64 s64;
  56 #ifdef __SSE2__
  57   __m128i m;
  58 #endif
  59   // this may be tempting, but it causes gcc to do lots of stack spills
  60   //gvhreg h[2];
  61 } gvreg;
  62
  63 typedef gvreg    gvreg_ua    __attribute__((aligned(1)));
  64 typedef uint64_t uint64_t_ua __attribute__((aligned(1)));
  65 typedef gvu8     gvu8_ua     __attribute__((aligned(1)));
  66 typedef gvu16    gvu16_ua    __attribute__((aligned(1)));
  67
  68 #if defined(__ARM_NEON) || defined(__ARM_NEON__)
  69 #include <arm_neon.h>
  70
  71 typedef union
  72 {
  73   gvhu8  u8;
  74   gvhu16 u16;
  75   gvhu32 u32;
  76   gvhu64 u64;
  77   //u64    u64;
  78   //uint64x1_t u64;
  79   gvhs8  s8;
  80   gvhs16 s16;
  81   gvhs32 s32;
  82   gvhs64 s64;
  83   //s64    s64;
  84   //int64x1_t s64;
  85 } gvhreg;
  86
  87 #define gvaddhn_u32(d, a, b)     d.u16 = vaddhn_u32(a.u32, b.u32)
  88 #define gvaddw_s32(d, a, b)      d.s64 = vaddw_s32(a.s64, b.s32)
  89 #define gvabsq_s32(d, s)         d.s32 = vabsq_s32(s.s32)
  90 #define gvbic_n_u16(d, n)        d.u16 = vbic_u16(d.u16, vmov_n_u16(n))
  91 #define gvbifq(d, a, b)          d.u8  = vbslq_u8(b.u8, d.u8, a.u8)
  92 #define gvbit(d, a, b)           d.u8  = vbsl_u8(b.u8, a.u8, d.u8)
  93 #define gvceqq_u16(d, a, b)      d.u16 = vceqq_u16(a.u16, b.u16)
  94 #define gvcgt_s16(d, a, b)       d.u16 = vcgt_s16(a.s16, b.s16)
  95 #define gvclt_s16(d, a, b)       d.u16 = vclt_s16(a.s16, b.s16)
  96 #define gvcreate_s32(d, a, b)    d.s32 = vcreate_s32((u32)(a) | ((u64)(b) << 32))
  97 #define gvcreate_u32(d, a, b)    d.u32 = vcreate_u32((u32)(a) | ((u64)(b) << 32))
  98 #define gvcreate_s64(d, s)       d.s64 = (gvhs64)vcreate_s64(s)
  99 #define gvcreate_u64(d, s)       d.u64 = (gvhu64)vcreate_u64(s)
 100 #define gvcombine_u16(d, l, h)   d.u16 = vcombine_u16(l.u16, h.u16)
 101 #define gvcombine_u32(d, l, h)   d.u32 = vcombine_u32(l.u32, h.u32)
 102 #define gvcombine_s64(d, l, h)   d.s64 = vcombine_s64((int64x1_t)l.s64, (int64x1_t)h.s64)
 103 #define gvdup_l_u8(d, s, l)      d.u8  = vdup_lane_u8(s.u8, l)
 104 #define gvdup_l_u16(d, s, l)     d.u16 = vdup_lane_u16(s.u16, l)
 105 #define gvdup_l_u32(d, s, l)     d.u32 = vdup_lane_u32(s.u32, l)
 106 #define gvdupq_l_s64(d, s, l)    d.s64 = vdupq_lane_s64((int64x1_t)s.s64, l)
 107 #define gvdupq_l_u32(d, s, l)    d.u32 = vdupq_lane_u32(s.u32, l)
 108 #define gvdup_n_s64(d, n)        d.s64 = vdup_n_s64(n)
 109 #define gvdup_n_u8(d, n)         d.u8  = vdup_n_u8(n)
 110 #define gvdup_n_u16(d, n)        d.u16 = vdup_n_u16(n)
 111 #define gvdup_n_u32(d, n)        d.u32 = vdup_n_u32(n)
 112 #define gvdupq_n_u16(d, n)       d.u16 = vdupq_n_u16(n)
 113 #define gvdupq_n_u32(d, n)       d.u32 = vdupq_n_u32(n)
 114 #define gvdupq_n_s64(d, n)       d.s64 = vdupq_n_s64(n)
 115 #define gvhaddq_u16(d, a, b)     d.u16 = vhaddq_u16(a.u16, b.u16)
 116 #define gvmax_s16(d, a, b)       d.s16 = vmax_s16(a.s16, b.s16)
 117 #define gvmin_s16(d, a, b)       d.s16 = vmin_s16(a.s16, b.s16)
 118 #define gvmin_u16(d, a, b)       d.u16 = vmin_u16(a.u16, b.u16)
 119 #define gvminq_u8(d, a, b)       d.u8  = vminq_u8(a.u8, b.u8)
 120 #define gvminq_u16(d, a, b)      d.u16 = vminq_u16(a.u16, b.u16)
 121 #define gvmla_s32(d, a, b)       d.s32 = vmla_s32(d.s32, a.s32, b.s32)
 122 #define gvmla_u32(d, a, b)       d.u32 = vmla_u32(d.u32, a.u32, b.u32)
 123 #define gvmlaq_s32(d, a, b)      d.s32 = vmlaq_s32(d.s32, a.s32, b.s32)
 124 #define gvmlaq_u32(d, a, b)      d.u32 = vmlaq_u32(d.u32, a.u32, b.u32)
 125 #define gvmlal_s32(d, a, b)      d.s64 = vmlal_s32(d.s64, a.s32, b.s32)
 126 #define gvmlal_u8(d, a, b)       d.u16 = vmlal_u8(d.u16, a.u8, b.u8)
 127 #define gvmlsq_s32(d, a, b)      d.s32 = vmlsq_s32(d.s32, a.s32, b.s32)
 128 #define gvmlsq_l_s32(d, a, b, l) d.s32 = vmlsq_lane_s32(d.s32, a.s32, b.s32, l)
 129 #define gvmov_l_s32(d, s, l)     d.s32 = vset_lane_s32(s, d.s32, l)
 130 #define gvmov_l_u32(d, s, l)     d.u32 = vset_lane_u32(s, d.u32, l)
 131 #define gvmovl_u8(d, s)          d.u16 = vmovl_u8(s.u8)
 132 #define gvmovl_s32(d, s)         d.s64 = vmovl_s32(s.s32)
 133 #define gvmovn_u16(d, s)         d.u8  = vmovn_u16(s.u16)
 134 #define gvmovn_u32(d, s)         d.u16 = vmovn_u32(s.u32)
 135 #define gvmovn_u64(d, s)         d.u32 = vmovn_u64(s.u64)
 136 #define gvmul_s32(d, a, b)       d.s32 = vmul_s32(a.s32, b.s32)
 137 #define gvmull_s16(d, a, b)      d.s32 = vmull_s16(a.s16, b.s16)
 138 #define gvmull_s32(d, a, b)      d.s64 = vmull_s32(a.s32, b.s32)
 139 #define gvmull_u8(d, a, b)       d.u16 = vmull_u8(a.u8, b.u8)
 140 #define gvmull_l_u32(d, a, b, l) d.u64 = vmull_lane_u32(a.u32, b.u32, l)
 141 #define gvmlsl_s16(d, a, b)      d.s32 = vmlsl_s16(d.s32, a.s16, b.s16)
 142 #define gvneg_s32(d, s)          d.s32 = vneg_s32(s.s32)
 143 #define gvqadd_u8(d, a, b)       d.u8  = vqadd_u8(a.u8, b.u8)
 144 #define gvqsub_u8(d, a, b)       d.u8  = vqsub_u8(a.u8, b.u8)
 145 #define gvshl_u16(d, a, b)       d.u16 = vshl_u16(a.u16, b.s16)
 146 #define gvshlq_u64(d, a, b)      d.u64 = vshlq_u64(a.u64, b.s64)
 147 #define gvshrq_n_s16(d, s, n)    d.s16 = vshrq_n_s16(s.s16, n)
 148 #define gvshrq_n_u16(d, s, n)    d.u16 = vshrq_n_u16(s.u16, n)
 149 #define gvshl_n_u32(d, s, n)     d.u32 = vshl_n_u32(s.u32, n)
 150 #define gvshlq_n_u16(d, s, n)    d.u16 = vshlq_n_u16(s.u16, n)
 151 #define gvshlq_n_u32(d, s, n)    d.u32 = vshlq_n_u32(s.u32, n)
 152 #define gvshll_n_s8(d, s, n)     d.s16 = vshll_n_s8(s.s8, n)
 153 #define gvshll_n_u8(d, s, n)     d.u16 = vshll_n_u8(s.u8, n)
 154 #define gvshll_n_u16(d, s, n)    d.u32 = vshll_n_u16(s.u16, n)
 155 #define gvshr_n_u8(d, s, n)      d.u8  = vshr_n_u8(s.u8, n)
 156 #define gvshr_n_u16(d, s, n)     d.u16 = vshr_n_u16(s.u16, n)
 157 #define gvshr_n_u32(d, s, n)     d.u32 = vshr_n_u32(s.u32, n)
 158 #define gvshr_n_u64(d, s, n)     d.u64 = (gvhu64)vshr_n_u64((uint64x1_t)s.u64, n)
 159 #define gvshrn_n_u16(d, s, n)    d.u8  = vshrn_n_u16(s.u16, n)
 160 #define gvshrn_n_u32(d, s, n)    d.u16 = vshrn_n_u32(s.u32, n)
 161 #define gvsli_n_u8(d, s, n)      d.u8  = vsli_n_u8(d.u8, s.u8, n)
 162 #define gvsri_n_u8(d, s, n)      d.u8  = vsri_n_u8(d.u8, s.u8, n)
 163 #define gvtstq_u16(d, a, b)      d.u16 = vtstq_u16(a.u16, b.u16)
 164 #define gvqshrun_n_s16(d, s, n)  d.u8  = vqshrun_n_s16(s.s16, n)
 165 #define gvqsubq_u8(d, a, b)      d.u8  = vqsubq_u8(a.u8, b.u8)
 166 #define gvqsubq_u16(d, a, b)     d.u16 = vqsubq_u16(a.u16, b.u16)
 167
 168 #define gvmovn_top_u64(d, s)     d.u32 = vshrn_n_u64(s.u64, 32)
 169
 170 #define gvget_lo(d, s)           d.u16 = vget_low_u16(s.u16)
 171 #define gvget_hi(d, s)           d.u16 = vget_high_u16(s.u16)
 172 #define gvlo(s)                  ({gvhreg t_; gvget_lo(t_, s); t_;})
 173 #define gvhi(s)                  ({gvhreg t_; gvget_hi(t_, s); t_;})
 174
 175 #define gvset_lo(d, s)           d.u16 = vcombine_u16(s.u16, gvhi(d).u16)
 176 #define gvset_hi(d, s)           d.u16 = vcombine_u16(gvlo(d).u16, s.u16)
 177
 178 #define gvtbl2_u8(d, a, b) { \
 179   uint8x8x2_t v_; \
 180   v_.val[0] = vget_low_u8(a.u8); v_.val[1] = vget_high_u8(a.u8); \
 181   d.u8 = vtbl2_u8(v_, b.u8); \
 182 }
 183
 184 #define gvzip_u8(d, a, b) { \
 185   uint8x8x2_t v_ = vzip_u8(a.u8, b.u8); \
 186   d.u8 = vcombine_u8(v_.val[0], v_.val[1]); \
 187 }
 188 #define gvzipq_u16(d0, d1, s0, s1) { \
 189   uint16x8x2_t v_ = vzipq_u16(s0.u16, s1.u16); \
 190   d0.u16 = v_.val[0]; d1.u16 = v_.val[1]; \
 191 }
 192
 193 #define gvld1_u8(d, s)           d.u8  = vld1_u8(s)
 194 #define gvld1_u32(d, s)          d.u32 = vld1_u32((const u32 *)(s))
 195 #define gvld1_u64(d, s)          d.u64 = vld1_u64((const u64 *)(s))
 196 #define gvld1q_u8(d, s)          d.u8  = vld1q_u8(s)
 197 #define gvld1q_u16(d, s)         d.u16 = vld1q_u16(s)
 198 #define gvld1q_u32(d, s)         d.u32 = vld1q_u32((const u32 *)(s))
 199 #define gvld2_u8_dup(v0, v1, p) { \
 200   uint8x8x2_t v_ = vld2_dup_u8(p); \
 201   v0.u8 = v_.val[0]; v1.u8 = v_.val[1]; \
 202 }
 203 #define gvld2q_u8(v0, v1, p) { \
 204   uint8x16x2_t v_ = vld2q_u8(p); \
 205   v0.u8 = v_.val[0]; v1.u8 = v_.val[1]; \
 206 }
 207
 208 #define gvst1_u8(v, p) \
 209   vst1_u8(p, v.u8)
 210 #define gvst1_u64(v, p) \
 211   vst1_u64((u64 *)(p), v.u64)
 212 #define gvst1q_u16(v, p) \
 213   vst1q_u16(p, v.u16)
 214 #define gvst1q_inc_u32(v, p, i) { \
 215   vst1q_u32((u32 *)(p), v.u32); \
 216   p += (i) / sizeof(*p); \
 217 }
 218 #define gvst2_u8(v0, v1, p) { \
 219   uint8x8x2_t v_; \
 220   v_.val[0] = v0.u8; v_.val[1] = v1.u8; \
 221   vst2_u8(p, v_); \
 222 }
 223 #define gvst2_u16(v0, v1, p) { \
 224   uint16x4x2_t v_; \
 225   v_.val[0] = v0.u16; v_.val[1] = v1.u16; \
 226   vst2_u16(p, v_); \
 227 }
 228 #define gvst2q_u8(v0, v1, p) { \
 229   uint8x16x2_t v_; \
 230   v_.val[0] = v0.u8; v_.val[1] = v1.u8; \
 231   vst2q_u8(p, v_); \
 232 }
 233 #define gvst4_4_inc_u32(v0, v1, v2, v3, p, i) { \
 234   uint32x2x4_t v_; \
 235   v_.val[0] = v0.u32; v_.val[1] = v1.u32; v_.val[2] = v2.u32; v_.val[3] = v3.u32; \
 236   vst4_u32(p, v_); p += (i) / sizeof(*p); \
 237 }
 238 #define gvst4_pi_u16(v0, v1, v2, v3, p) { \
 239   uint16x4x4_t v_; \
 240   v_.val[0] = v0.u16; v_.val[1] = v1.u16; v_.val[2] = v2.u16; v_.val[3] = v3.u16; \
 241   vst4_u16((u16 *)(p), v_); p += sizeof(v_) / sizeof(*p); \
 242 }
 243 #define gvst1q_pi_u32(v, p) \
 244   gvst1q_inc_u32(v, p, sizeof(v))
 245 // could use vst1q_u32_x2 but that's not always available
 246 #define gvst1q_2_pi_u32(v0, v1, p) { \
 247   gvst1q_inc_u32(v0, p, sizeof(v0)); \
 248   gvst1q_inc_u32(v1, p, sizeof(v1)); \
 249 }
 250
 251 /* notes:
 252  - gcc > 9: (arm32) int64x1_t type produces ops on gp regs
 253             (also u64 __attribute__((vector_size(8))) :( )
 254  - gcc <11: (arm32) handles '<vec> == 0' poorly
 255 */
 256
 257 #elif defined(__SSE2__)
 258
 259 // use a full reg and discard the upper half
 260 #define gvhreg gvreg
 261
 262 #define gv0()                    _mm_setzero_si128()
 263
 264 #ifdef __x86_64__
 265 #define gvcreate_s32(d, a, b)    d.m = _mm_cvtsi64_si128((u32)(a) | ((u64)(b) << 32))
 266 #define gvcreate_s64(d, s)       d.m = _mm_cvtsi64_si128(s)
 267 #else
 268 #define gvcreate_s32(d, a, b)    d.m = _mm_set_epi32(0, 0, b, a)
 269 #define gvcreate_s64(d, s)       d.m = _mm_loadu_si64(&(s))
 270 #endif
 271
 272 #define gvbic_n_u16(d, n)        d.m = _mm_andnot_si128(_mm_set1_epi16(n), d.m)
 273 #define gvceqq_u16(d, a, b)      d.u16 = vceqq_u16(a.u16, b.u16)
 274 #define gvcgt_s16(d, a, b)       d.m = _mm_cmpgt_epi16(a.m, b.m)
 275 #define gvclt_s16(d, a, b)       d.m = _mm_cmpgt_epi16(b.m, a.m)
 276 #define gvcreate_u32             gvcreate_s32
 277 #define gvcreate_u64             gvcreate_s64
 278 #define gvcombine_u16(d, l, h)   d.m = _mm_unpacklo_epi64(l.m, h.m)
 279 #define gvcombine_u32            gvcombine_u16
 280 #define gvcombine_s64            gvcombine_u16
 281 #define gvdup_l_u8(d, s, l)      d.u8  = vdup_lane_u8(s.u8, l)
 282 #define gvdup_l_u16(d, s, l)     d.m = _mm_shufflelo_epi16(s.m, (l)|((l)<<2)|((l)<<4)|((l)<<6))
 283 #define gvdup_l_u32(d, s, l)     d.m =  vdup_lane_u32(s.u32, l)
 284 #define gvdupq_l_s64(d, s, l)    d.m = _mm_unpacklo_epi64(s.m, s.m)
 285 #define gvdupq_l_u32(d, s, l)    d.m = _mm_shuffle_epi32(s.m, (l)|((l)<<2)|((l)<<4)|((l)<<6))
 286 #define gvdup_n_s64(d, n)        d.m = _mm_set1_epi64x(n)
 287 #define gvdup_n_u8(d, n)         d.m = _mm_set1_epi8(n)
 288 #define gvdup_n_u16(d, n)        d.m = _mm_set1_epi16(n)
 289 #define gvdup_n_u32(d, n)        d.m = _mm_set1_epi32(n)
 290 #define gvdupq_n_u16(d, n)       d.m = _mm_set1_epi16(n)
 291 #define gvdupq_n_u32(d, n)       d.m = _mm_set1_epi32(n)
 292 #define gvdupq_n_s64(d, n)       d.m = _mm_set1_epi64x(n)
 293 #define gvmax_s16(d, a, b)       d.m = _mm_max_epi16(a.m, b.m)
 294 #define gvmin_s16(d, a, b)       d.m = _mm_min_epi16(a.m, b.m)
 295 #define gvminq_u8(d, a, b)       d.m = _mm_min_epu8(a.m, b.m)
 296 #define gvmovn_u64(d, s)         d.m = _mm_shuffle_epi32(s.m, 0 | (2 << 2))
 297 #define gvmovn_top_u64(d, s)     d.m = _mm_shuffle_epi32(s.m, 1 | (3 << 2))
 298 #define gvmull_s16(d, a, b) { \
 299   __m128i lo_ = _mm_mullo_epi16(a.m, b.m); \
 300   __m128i hi_ = _mm_mulhi_epi16(a.m, b.m); \
 301   d.m = _mm_unpacklo_epi16(lo_, hi_); \
 302 }
 303 #define gvmull_l_u32(d, a, b, l) { \
 304   __m128i a_ = _mm_unpacklo_epi32(a.m, a.m); /* lanes 0,1 -> 0,2 */ \
 305   __m128i b_ = _mm_shuffle_epi32(b.m, (l) | ((l) << 4)); \
 306   d.m = _mm_mul_epu32(a_, b_); \
 307 }
 308 #define gvmlsl_s16(d, a, b) { \
 309   gvreg tmp_; \
 310   gvmull_s16(tmp_, a, b); \
 311   d.m = _mm_sub_epi32(d.m, tmp_.m); \
 312 }
 313 #define gvqadd_u8(d, a, b)       d.m = _mm_adds_epu8(a.m, b.m)
 314 #define gvqsub_u8(d, a, b)       d.m = _mm_subs_epu8(a.m, b.m)
 315 #define gvshrq_n_s16(d, s, n)    d.m = _mm_srai_epi16(s.m, n)
 316 #define gvshrq_n_u16(d, s, n)    d.m = _mm_srli_epi16(s.m, n)
 317 #define gvshrq_n_u32(d, s, n)    d.m = _mm_srli_epi32(s.m, n)
 318 #define gvshl_n_u32(d, s, n)     d.m = _mm_slli_epi32(s.m, n)
 319 #define gvshlq_n_u16(d, s, n)    d.m = _mm_slli_epi16(s.m, n)
 320 #define gvshlq_n_u32(d, s, n)    d.m = _mm_slli_epi32(s.m, n)
 321 #define gvshll_n_u16(d, s, n)    d.m = _mm_slli_epi32(_mm_unpacklo_epi16(s.m, gv0()), n)
 322 #define gvshr_n_u16(d, s, n)     d.m = _mm_srli_epi16(s.m, n)
 323 #define gvshr_n_u32(d, s, n)     d.m = _mm_srli_epi32(s.m, n)
 324 #define gvshr_n_u64(d, s, n)     d.m = _mm_srli_epi64(s.m, n)
 325 #define gvshrn_n_s64(d, s, n) { \
 326   gvreg tmp_; \
 327   gvshrq_n_s64(tmp_, s, n); \
 328   d.m = _mm_shuffle_epi32(tmp_.m, 0 | (2 << 2)); \
 329 }
 330 #define gvqshrun_n_s16(d, s, n) { \
 331   __m128i t_ = _mm_srai_epi16(s.m, n); \
 332   d.m = _mm_packus_epi16(t_, t_); \
 333 }
 334 #define gvqsubq_u8(d, a, b)      d.m = _mm_subs_epu8(a.m, b.m)
 335 #define gvqsubq_u16(d, a, b)     d.m = _mm_subs_epu16(a.m, b.m)
 336
 337 #ifdef __SSSE3__
 338 #define gvabsq_s32(d, s)         d.m = _mm_abs_epi32(s.m)
 339 #define gvtbl2_u8(d, a, b)       d.m = _mm_shuffle_epi8(a.m, b.m)
 340 #else
 341 // must supply these here or else gcc will produce something terrible with __builtin_shuffle
 342 #define gvmovn_u16(d, s) { \
 343   __m128i t2_ = _mm_and_si128(s.m, _mm_set1_epi16(0xff)); \
 344   d.m = _mm_packus_epi16(t2_, t2_); \
 345 }
 346 #define gvmovn_u32(d, s) { \
 347   __m128i t2_; \
 348   t2_ = _mm_shufflelo_epi16(s.m, (0 << 0) | (2 << 2)); \
 349   t2_ = _mm_shufflehi_epi16(t2_, (0 << 0) | (2 << 2)); \
 350   d.m = _mm_shuffle_epi32(t2_, (0 << 0) | (2 << 2)); \
 351 }
 352 #define gvmovn_top_u32(d, s) { \
 353   __m128i t2_; \
 354   t2_ = _mm_shufflelo_epi16(s.m, (1 << 0) | (3 << 2)); \
 355   t2_ = _mm_shufflehi_epi16(t2_, (1 << 0) | (3 << 2)); \
 356   d.m = _mm_shuffle_epi32(t2_, (0 << 0) | (2 << 2)); \
 357 }
 358 #endif // !__SSSE3__
 359 #ifdef __SSE4_1__
 360 #define gvmin_u16(d, a, b)       d.m = _mm_min_epu16(a.m, b.m)
 361 #define gvminq_u16               gvmin_u16
 362 #define gvmovl_u8(d, s)          d.m = _mm_cvtepu8_epi16(s.m)
 363 #define gvmovl_s8(d, s)          d.m = _mm_cvtepi8_epi16(s.m)
 364 #define gvmovl_s32(d, s)         d.m = _mm_cvtepi32_epi64(s.m)
 365 #define gvmull_s32(d, a, b) { \
 366   __m128i a_ = _mm_unpacklo_epi32(a.m, a.m); /* lanes 0,1 -> 0,2 */ \
 367   __m128i b_ = _mm_unpacklo_epi32(b.m, b.m); \
 368   d.m = _mm_mul_epi32(a_, b_); \
 369 }
 370 #else
 371 #define gvmovl_u8(d, s)          d.m = _mm_unpacklo_epi8(s.m, gv0())
 372 #define gvmovl_s8(d, s)          d.m = _mm_unpacklo_epi8(s.m, _mm_cmpgt_epi8(gv0(), s.m))
 373 #define gvmovl_s32(d, s)         d.m = _mm_unpacklo_epi32(s.m, _mm_srai_epi32(s.m, 31))
 374 #endif // !__SSE4_1__
 375 #ifndef __AVX2__
 376 #define gvshlq_u64(d, a, b) { \
 377   gvreg t1_, t2_; \
 378   t1_.m = _mm_sll_epi64(a.m, b.m); \
 379   t2_.m = _mm_sll_epi64(a.m, _mm_shuffle_epi32(b.m, (2 << 0) | (3 << 2))); \
 380   d.u64 = (gvu64){ t1_.u64[0], t2_.u64[1] }; \
 381 }
 382 #endif // __AVX2__
 383
 384 #define gvlo(s)                  s
 385 #define gvhi(s)                  ((gvreg)_mm_shuffle_epi32(s.m, (2 << 0) | (3 << 2)))
 386 #define gvget_lo(d, s)           d = gvlo(s)
 387 #define gvget_hi(d, s)           d = gvhi(s)
 388
 389 #define gvset_lo(d, s)           d.m = _mm_unpacklo_epi64(s.m, gvhi(d).m)
 390 #define gvset_hi(d, s)           d.m = _mm_unpacklo_epi64(d.m, s.m)
 391
 392 #define gvld1_u8(d, s)           d.m = _mm_loadu_si64(s)
 393 #define gvld1_u32                gvld1_u8
 394 #define gvld1_u64                gvld1_u8
 395 #define gvld1q_u8(d, s)          d.m = _mm_loadu_si128((__m128i *)(s))
 396 #define gvld1q_u16               gvld1q_u8
 397 #define gvld1q_u32               gvld1q_u8
 398
 399 #define gvst1_u8(v, p)           _mm_storeu_si64(p, v.m)
 400 #define gvst1_u64                gvst1_u8
 401
 402 #define gvst4_4_inc_u32(v0, v1, v2, v3, p, i) { \
 403   __m128i t0 = _mm_unpacklo_epi32(v0.m, v1.m); \
 404   __m128i t1 = _mm_unpacklo_epi32(v2.m, v3.m); \
 405   _mm_storeu_si128(((__m128i *)(p)) + 0, _mm_unpacklo_epi64(t0, t1)); \
 406   _mm_storeu_si128(((__m128i *)(p)) + 1, _mm_unpackhi_epi64(t0, t1)); \
 407   p += (i) / sizeof(*p); \
 408 }
 409 #define gvst4_pi_u16(v0, v1, v2, v3, p) { \
 410   __m128i t0 = _mm_unpacklo_epi16(v0.m, v1.m); \
 411   __m128i t1 = _mm_unpacklo_epi16(v2.m, v3.m); \
 412   _mm_storeu_si128(((__m128i *)(p)) + 0, _mm_unpacklo_epi32(t0, t1)); \
 413   _mm_storeu_si128(((__m128i *)(p)) + 1, _mm_unpackhi_epi32(t0, t1)); \
 414   p += sizeof(t0) * 2 / sizeof(*p); \
 415 }
 416
 417 #else
 418 #error "arch not supported or SIMD support was not enabled by your compiler"
 419 #endif
 420
 421 // the below have intrinsics but they evaluate to basic operations on both gcc and clang
 422 #define gvadd_s64(d, a, b)       d.s64 = a.s64 + b.s64
 423 #define gvadd_u8(d, a, b)        d.u8  = a.u8  + b.u8
 424 #define gvadd_u16(d, a, b)       d.u16 = a.u16 + b.u16
 425 #define gvadd_u32(d, a, b)       d.u32 = a.u32 + b.u32
 426 #define gvaddq_s64               gvadd_s64
 427 #define gvaddq_u16               gvadd_u16
 428 #define gvaddq_u32               gvadd_u32
 429 #define gvand(d, a, b)           d.u32 = a.u32 & b.u32
 430 #define gvand_n_u32(d, n)        d.u32 &= n
 431 #define gvbic(d, a, b)           d.u32 = a.u32 & ~b.u32
 432 #define gvbicq                   gvbic
 433 #define gveor(d, a, b)           d.u32 = a.u32 ^ b.u32
 434 #define gveorq                   gveor
 435 #define gvceqz_u16(d, s)         d.u16 = s.u16 == 0
 436 #define gvceqzq_u16              gvceqz_u16
 437 #define gvcltz_s16(d, s)         d.s16 = s.s16 < 0
 438 #define gvcltzq_s16              gvcltz_s16
 439 #define gvsub_u16(d, a, b)       d.u16 = a.u16 - b.u16
 440 #define gvsub_u32(d, a, b)       d.u32 = a.u32 - b.u32
 441 #define gvsubq_u16               gvsub_u16
 442 #define gvsubq_u32               gvsub_u32
 443 #define gvorr(d, a, b)           d.u32 = a.u32 | b.u32
 444 #define gvorrq                   gvorr
 445 #define gvorr_n_u16(d, n)        d.u16 |= n
 446
 447 // fallbacks
 448 #if 1
 449
 450 #ifndef gvaddhn_u32
 451 #define gvaddhn_u32(d, a, b) { \
 452   gvreg tmp1_ = { .u32 = a.u32 + b.u32 }; \
 453   gvmovn_top_u32(d, tmp1_); \
 454 }
 455 #endif
 456 #ifndef gvabsq_s32
 457 #define gvabsq_s32(d, s) { \
 458   gvreg tmp1_ = { .s32 = (gvs32){} - s.s32 }; \
 459   gvreg mask_ = { .s32 = s.s32 >> 31 }; \
 460   gvbslq_(d, mask_, tmp1_, s); \
 461 }
 462 #endif
 463 #ifndef gvbit
 464 #define gvbslq_(d, s, a, b)      d.u32 = (a.u32 & s.u32) | (b.u32 & ~s.u32)
 465 #define gvbifq(d, a, b)          gvbslq_(d, b, d, a)
 466 #define gvbit(d, a, b)           gvbslq_(d, b, a, d)
 467 #endif
 468 #ifndef gvaddw_s32
 469 #define gvaddw_s32(d, a, b)     {gvreg t_; gvmovl_s32(t_, b); d.s64 += t_.s64;}
 470 #endif
 471 #ifndef gvhaddq_u16
 472 // can do this because the caller needs the msb clear
 473 #define gvhaddq_u16(d, a, b)     d.u16 = (a.u16 + b.u16) >> 1
 474 #endif
 475 #ifndef gvmin_u16
 476 #define gvmin_u16(d, a, b) { \
 477   gvu16 t_ = a.u16 < b.u16; \
 478   d.u16 = (a.u16 & t_) | (b.u16 & ~t_); \
 479 }
 480 #define gvminq_u16               gvmin_u16
 481 #endif
 482 #ifndef gvmlsq_s32
 483 #define gvmlsq_s32(d, a, b)      d.s32 -= a.s32 * b.s32
 484 #endif
 485 #ifndef gvmlsq_l_s32
 486 #define gvmlsq_l_s32(d, a, b, l){gvreg t_; gvdupq_l_u32(t_, b, l); d.s32 -= a.s32 * t_.s32;}
 487 #endif
 488 #ifndef gvmla_s32
 489 #define gvmla_s32(d, a, b)       d.s32 += a.s32 * b.s32
 490 #endif
 491 #ifndef gvmla_u32
 492 #define gvmla_u32                gvmla_s32
 493 #endif
 494 #ifndef gvmlaq_s32
 495 #define gvmlaq_s32(d, a, b)      d.s32 += a.s32 * b.s32
 496 #endif
 497 #ifndef gvmlaq_u32
 498 #define gvmlaq_u32               gvmlaq_s32
 499 #endif
 500 #ifndef gvmlal_u8
 501 #define gvmlal_u8(d, a, b)      {gvreg t_; gvmull_u8(t_, a, b); d.u16 += t_.u16;}
 502 #endif
 503 #ifndef gvmlal_s32
 504 #define gvmlal_s32(d, a, b)     {gvreg t_; gvmull_s32(t_, a, b); d.s64 += t_.s64;}
 505 #endif
 506 #ifndef gvmov_l_s32
 507 #define gvmov_l_s32(d, s, l)     d.s32[l] = s
 508 #endif
 509 #ifndef gvmov_l_u32
 510 #define gvmov_l_u32(d, s, l)     d.u32[l] = s
 511 #endif
 512 #ifndef gvmul_s32
 513 #define gvmul_s32(d, a, b)       d.s32 = a.s32 * b.s32
 514 #endif
 515 #ifndef gvmull_u8
 516 #define gvmull_u8(d, a, b) { \
 517   gvreg t1_, t2_; \
 518   gvmovl_u8(t1_, a); \
 519   gvmovl_u8(t2_, b); \
 520   d.u16 = t1_.u16 * t2_.u16; \
 521 }
 522 #endif
 523 #ifndef gvmull_s32
 524 // note: compilers tend to use int regs here
 525 #define gvmull_s32(d, a, b) { \
 526   d.s64[0] = (s64)a.s32[0] * b.s32[0]; \
 527   d.s64[1] = (s64)a.s32[1] * b.s32[1]; \
 528 }
 529 #endif
 530 #ifndef gvneg_s32
 531 #define gvneg_s32(d, s)          d.s32 = -s.s32
 532 #endif
 533 // x86 note: needs _mm_sllv_epi16 (avx512), else this sucks terribly
 534 #ifndef gvshl_u16
 535 #define gvshl_u16(d, a, b)       d.u16 = a.u16 << b.u16
 536 #endif
 537 // x86 note: needs _mm_sllv_* (avx2)
 538 #ifndef gvshlq_u64
 539 #define gvshlq_u64(d, a, b)      d.u64 = a.u64 << b.u64
 540 #endif
 541 #ifndef gvshll_n_s8
 542 #define gvshll_n_s8(d, s, n)    {gvreg t_; gvmovl_s8(t_, s); gvshlq_n_u16(d, t_, n);}
 543 #endif
 544 #ifndef gvshll_n_u8
 545 #define gvshll_n_u8(d, s, n)    {gvreg t_; gvmovl_u8(t_, s); gvshlq_n_u16(d, t_, n);}
 546 #endif
 547 #ifndef gvshr_n_u8
 548 #define gvshr_n_u8(d, s, n)      d.u8  = s.u8 >> (n)
 549 #endif
 550 #ifndef gvshrq_n_s64
 551 #define gvshrq_n_s64(d, s, n)    d.s64 = s.s64 >> (n)
 552 #endif
 553 #ifndef gvshrn_n_u16
 554 #define gvshrn_n_u16(d, s, n)   {gvreg t_; gvshrq_n_u16(t_, s, n); gvmovn_u16(d, t_);}
 555 #endif
 556 #ifndef gvshrn_n_u32
 557 #define gvshrn_n_u32(d, s, n)   {gvreg t_; gvshrq_n_u32(t_, s, n); gvmovn_u32(d, t_);}
 558 #endif
 559 #ifndef gvsli_n_u8
 560 #define gvsli_n_u8(d, s, n)      d.u8 = (s.u8 << (n)) | (d.u8 & ((1u << (n)) - 1u))
 561 #endif
 562 #ifndef gvsri_n_u8
 563 #define gvsri_n_u8(d, s, n)      d.u8 = (s.u8 >> (n)) | (d.u8 & ((0xff00u >> (n)) & 0xffu))
 564 #endif
 565 #ifndef gvtstq_u16
 566 #define gvtstq_u16(d, a, b)      d.u16 = (a.u16 & b.u16) != 0
 567 #endif
 568
 569 #ifndef gvld2_u8_dup
 570 #define gvld2_u8_dup(v0, v1, p) { \
 571   gvdup_n_u8(v0, ((const u8 *)(p))[0]); \
 572   gvdup_n_u8(v1, ((const u8 *)(p))[1]); \
 573 }
 574 #endif
 575 #ifndef gvst1_u8
 576 #define gvst1_u8(v, p)           *(uint64_t_ua *)(p) = v.u64[0]
 577 #endif
 578 #ifndef gvst1q_u16
 579 #define gvst1q_u16(v, p)         *(gvreg_ua *)(p) = v
 580 #endif
 581 #ifndef gvst1q_inc_u32
 582 #define gvst1q_inc_u32(v, p, i) {*(gvreg_ua *)(p) = v; p += (i) / sizeof(*p);}
 583 #endif
 584 #ifndef gvst1q_pi_u32
 585 #define gvst1q_pi_u32(v, p)      gvst1q_inc_u32(v, p, sizeof(v))
 586 #endif
 587 #ifndef gvst1q_2_pi_u32
 588 #define gvst1q_2_pi_u32(v0, v1, p) { \
 589   gvst1q_inc_u32(v0, p, sizeof(v0)); \
 590   gvst1q_inc_u32(v1, p, sizeof(v1)); \
 591 }
 592 #endif
 593 #ifndef gvst2_u8
 594 #define gvst2_u8(v0, v1, p)     {gvreg t_; gvzip_u8(t_, v0, v1); *(gvu8_ua *)(p) = t_.u8;}
 595 #endif
 596 #ifndef gvst2_u16
 597 #define gvst2_u16(v0, v1, p)    {gvreg t_; gvzip_u16(t_, v0, v1); *(gvu16_ua *)(p) = t_.u16;}
 598 #endif
 599
 600 // note: these shuffles assume sizeof(gvhreg) == 16 && sizeof(gvreg) == 16
 601 #ifndef __has_builtin
 602 #define __has_builtin(x) 0
 603 #endif
 604
 605 // prefer __builtin_shuffle on gcc as it handles -1 poorly
 606 #if __has_builtin(__builtin_shufflevector) && !__has_builtin(__builtin_shuffle)
 607
 608 #ifndef gvld2q_u8
 609 #define gvld2q_u8(v0, v1, p) { \
 610   gvu8 v0_ = ((gvu8_ua *)(p))[0]; \
 611   gvu8 v1_ = ((gvu8_ua *)(p))[1]; \
 612   v0.u8 = __builtin_shufflevector(v0_, v1_, 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30); \
 613   v1.u8 = __builtin_shufflevector(v0_, v1_, 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31); \
 614 }
 615 #endif
 616 #ifndef gvmovn_u16
 617 #define gvmovn_u16(d, s) \
 618   d.u8 = __builtin_shufflevector(s.u8, s.u8, 0,2,4,6,8,10,12,14,-1,-1,-1,-1,-1,-1,-1,-1)
 619 #endif
 620 #ifndef gvmovn_u32
 621 #define gvmovn_u32(d, s) \
 622   d.u16 = __builtin_shufflevector(s.u16, s.u16, 0,2,4,6,-1,-1,-1,-1)
 623 #endif
 624 #ifndef gvmovn_top_u32
 625 #define gvmovn_top_u32(d, s) \
 626   d.u16 = __builtin_shufflevector(s.u16, s.u16, 1,3,5,7,-1,-1,-1,-1)
 627 #endif
 628 #ifndef gvzip_u8
 629 #define gvzip_u8(d, a, b) \
 630   d.u8 = __builtin_shufflevector(a.u8, b.u8, 0,16,1,17,2,18,3,19,4,20,5,21,6,22,7,23)
 631 #endif
 632 #ifndef gvzip_u16
 633 #define gvzip_u16(d, a, b) \
 634   d.u16 = __builtin_shufflevector(a.u16, b.u16, 0,8,1,9,2,10,3,11)
 635 #endif
 636 #ifndef gvzipq_u16
 637 #define gvzipq_u16(d0, d1, s0, s1) { \
 638   gvu16 t_ = __builtin_shufflevector(s0.u16, s1.u16, 0, 8, 1, 9, 2, 10, 3, 11); \
 639   d1.u16   = __builtin_shufflevector(s0.u16, s1.u16, 4,12, 5,13, 6, 14, 7, 15); \
 640   d0.u16 = t_; \
 641 }
 642 #endif
 643
 644 #else // !__has_builtin(__builtin_shufflevector)
 645
 646 #ifndef gvld2q_u8
 647 #define gvld2q_u8(v0, v1, p) { \
 648   gvu8 v0_ = ((gvu8_ua *)(p))[0]; \
 649   gvu8 v1_ = ((gvu8_ua *)(p))[1]; \
 650   v0.u8 = __builtin_shuffle(v0_, v1_, (gvu8){0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30}); \
 651   v1.u8 = __builtin_shuffle(v0_, v1_, (gvu8){1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31}); \
 652 }
 653 #endif
 654 #ifndef gvmovn_u16
 655 #define gvmovn_u16(d, s) \
 656   d.u8 = __builtin_shuffle(s.u8, (gvu8){0,2,4,6,8,10,12,14,0,2,4,6,8,10,12,14})
 657 #endif
 658 #ifndef gvmovn_u32
 659 #define gvmovn_u32(d, s) \
 660   d.u16 = __builtin_shuffle(s.u16, (gvu16){0,2,4,6,0,2,4,6})
 661 #endif
 662 #ifndef gvmovn_top_u32
 663 #define gvmovn_top_u32(d, s) \
 664   d.u16 = __builtin_shuffle(s.u16, (gvu16){1,3,5,7,1,3,5,7})
 665 #endif
 666 #ifndef gvtbl2_u8
 667 #define gvtbl2_u8(d, a, b)  d.u8 = __builtin_shuffle(a.u8, b.u8)
 668 #endif
 669 #ifndef gvzip_u8
 670 #define gvzip_u8(d, a, b) \
 671   d.u8 = __builtin_shuffle(a.u8, b.u8, (gvu8){0,16,1,17,2,18,3,19,4,20,5,21,6,22,7,23})
 672 #endif
 673 #ifndef gvzip_u16
 674 #define gvzip_u16(d, a, b) \
 675   d.u16 = __builtin_shuffle(a.u16, b.u16, (gvu16){0,8,1,9,2,10,3,11})
 676 #endif
 677 #ifndef gvzipq_u16
 678 #define gvzipq_u16(d0, d1, s0, s1) { \
 679   gvu16 t_ = __builtin_shuffle(s0.u16, s1.u16, (gvu16){0, 8, 1, 9, 2, 10, 3, 11}); \
 680   d1.u16   = __builtin_shuffle(s0.u16, s1.u16, (gvu16){4,12, 5,13, 6, 14, 7, 15}); \
 681   d0.u16 = t_; \
 682 }
 683 #endif
 684
 685 #endif // __builtin_shufflevector || __builtin_shuffle
 686
 687 #ifndef gvtbl2_u8
 688 #define gvtbl2_u8(d, a, b) { \
 689   int i_; \
 690   for (i_ = 0; i_ < 16; i_++) \
 691     d.u8[i_] = a.u8[b.u8[i_]]; \
 692 }
 693 #endif
 694
 695 #endif // fallbacks
 696
 697 #if defined(__arm__)
 698
 699 #define gssub16(d, a, b)    asm("ssub16 %0,%1,%2" : "=r"(d) : "r"(a), "r"(b))
 700 #define gsmusdx(d, a, b)    asm("smusdx %0,%1,%2" : "=r"(d) : "r"(a), "r"(b))
 701
 702 #if 0
 703 // gcc/config/arm/arm.c
 704 #undef gvadd_s64
 705 #define gvadd_s64(d, a, b)  asm("vadd.i64 %P0,%P1,%P2" : "=w"(d.s64) : "w"(a.s64), "w"(b.s64))
 706 #endif
 707
 708 #else
 709
 710 #define gssub16(d, a, b)    d = (u16)((a) - (b)) | ((((a) >> 16) - ((b) >> 16)) << 16)
 711 #define gsmusdx(d, a, b)    d = ((s32)(s16)(a) * ((s32)(b) >> 16)) \
 712                               - (((s32)(a) >> 16) * (s16)(b))
 713
 714 #endif
 715
 716 // for compatibility with the original psx_gpu.c code
 717 #define vec_2x64s gvreg
 718 #define vec_2x64u gvreg
 719 #define vec_4x32s gvreg
 720 #define vec_4x32u gvreg
 721 #define vec_8x16s gvreg
 722 #define vec_8x16u gvreg
 723 #define vec_16x8s gvreg
 724 #define vec_16x8u gvreg
 725 #define vec_1x64s gvhreg
 726 #define vec_1x64u gvhreg
 727 #define vec_2x32s gvhreg
 728 #define vec_2x32u gvhreg
 729 #define vec_4x16s gvhreg
 730 #define vec_4x16u gvhreg
 731 #define vec_8x8s  gvhreg
 732 #define vec_8x8u  gvhreg
 733
 734 #if 0
 735 #include <stdio.h>
 736 #include <stdlib.h>
 737 #include <unistd.h>
 738 static int ccount, dump_enabled;
 739 void cmpp(const char *name, const void *a_, const void *b_, size_t len)
 740 {
 741   const uint32_t *a = a_, *b = b_, masks[] = { 0, 0xff, 0xffff, 0xffffff };
 742   size_t i, left;
 743   uint32_t mask;
 744   for (i = 0; i < (len + 3)/4; i++) {
 745     left = len - i*4;
 746     mask = left >= 4 ? ~0u : masks[left];
 747     if ((a[i] ^ b[i]) & mask) {
 748       printf("%s: %08x %08x [%03zx/%zu] #%d\n",
 749         name, a[i] & mask, b[i] & mask, i*4, i, ccount);
 750       exit(1);
 751     }
 752   }
 753   ccount++;
 754 }
 755 #define ccmpf(n)   cmpp(#n, &psx_gpu->n, &n##_c, sizeof(n##_c))
 756 #define ccmpa(n,c) cmpp(#n, &psx_gpu->n, &n##_c, sizeof(n##_c[0]) * c)
 757
 758 void dump_r_(const char *name, void *dump, int is_q)
 759 {
 760   unsigned long long *u = dump;
 761   if (!dump_enabled) return;
 762   //if (ccount > 1) return;
 763   printf("%20s %016llx ", name, u[0]);
 764   if (is_q)
 765     printf("%016llx", u[1]);
 766   puts("");
 767 }
 768 void __attribute__((noinline,noclone)) dump_r_d(const char *name, void *dump)
 769 { dump_r_(name, dump, 0); }
 770 void __attribute__((noinline,noclone)) dump_r_q(const char *name, void *dump)
 771 { dump_r_(name, dump, 1); }
 772 #define dumprd(n) { u8 dump_[8]; gvst1_u8(n, dump_); dump_r_d(#n, dump_); }
 773 #define dumprq(n) { u16 dump_[8]; gvst1q_u16(n, dump_); dump_r_q(#n, dump_); }
 774 #endif
 775
 776 void compute_all_gradients(psx_gpu_struct * __restrict__ psx_gpu,
 777  const vertex_struct * __restrict__ a, const vertex_struct * __restrict__ b,
 778  const vertex_struct * __restrict__ c)
 779 {
 780   union { double d; struct { u32 l; u32 h; } i; } divident, divider;
 781   union { double d; gvhreg v; } d30;
 782
 783 #if 0
 784  compute_all_gradients_(psx_gpu, a, b, c);
 785  return;
 786 #endif
 787   // First compute the triangle area reciprocal and shift. The division will
 788   // happen concurrently with much of the work which follows.
 789
 790   // load exponent of 62 into upper half of double
 791   u32 shift = __builtin_clz(psx_gpu->triangle_area);
 792   u32 triangle_area_normalized = psx_gpu->triangle_area << shift;
 793
 794   // load area normalized into lower half of double
 795   divident.i.l = triangle_area_normalized >> 10;
 796   divident.i.h = (62 + 1023) << 20;
 797
 798   divider.i.l = triangle_area_normalized << 20;
 799   divider.i.h = ((1022 + 31) << 20) + (triangle_area_normalized >> 11);
 800
 801   d30.d = divident.d / divider.d;       // d30 = ((1 << 62) + ta_n) / ta_n
 802
 803   // ((x1 - x0) * (y2 - y1)) - ((x2 - x1) * (y1 - y0)) =
 804   // ( d0       *  d1      ) - ( d2       *  d3      ) =
 805   // ( m0                  ) - ( m1                  ) = gradient
 806
 807   // This is split to do 12 elements at a time over three sets: a, b, and c.
 808   // Technically we only need to do 10 elements (uvrgb_x and uvrgb_y), so
 809   // two of the slots are unused.
 810
 811   // Inputs are all 16-bit signed. The m0/m1 results are 32-bit signed, as
 812   // is g.
 813
 814   // First type is:  uvrg bxxx xxxx
 815   // Second type is: yyyy ybyy uvrg
 816   // Since x_a and y_c are the same the same variable is used for both.
 817
 818   gvreg v0;
 819   gvreg v1;
 820   gvreg v2;
 821   gvreg uvrg_xxxx0;
 822   gvreg uvrg_xxxx1;
 823   gvreg uvrg_xxxx2;
 824
 825   gvreg y0_ab;
 826   gvreg y1_ab;
 827   gvreg y2_ab;
 828
 829   gvreg d0_ab;
 830   gvreg d1_ab;
 831   gvreg d2_ab;
 832   gvreg d3_ab;
 833
 834   gvreg ga_uvrg_x;
 835   gvreg ga_uvrg_y;
 836   gvreg gw_rg_x;
 837   gvreg gw_rg_y;
 838   gvreg w_mask;
 839   gvreg r_shift;
 840   gvreg uvrg_dx2, uvrg_dx3;
 841   gvreg uvrgb_phase;
 842   gvhreg zero, tmp_lo, tmp_hi;
 843
 844   gvld1q_u8(v0, (u8 *)a);               // v0 = { uvrg0, b0, x0, y0 }
 845   gvld1q_u8(v1, (u8 *)b);               // v1 = { uvrg1, b1, x1, y1 }
 846   gvld1q_u8(v2, (u8 *)c);               // v2 = { uvrg2, b2, x2, y2 }
 847
 848   gvmovl_u8(uvrg_xxxx0, gvlo(v0));      // uvrg_xxxx0 = { uv0, rg0, b0-, -- }
 849   gvmovl_u8(uvrg_xxxx1, gvlo(v1));      // uvrg_xxxx1 = { uv1, rg1, b1-, -- }
 850   gvmovl_u8(uvrg_xxxx2, gvlo(v2));      // uvrg_xxxx2 = { uv2, rg2, b2-, -- }
 851
 852   gvdup_l_u16(tmp_lo, gvhi(v0), 1);     // yyyy0 = { yy0, yy0 }
 853   gvcombine_u16(y0_ab, tmp_lo, gvlo(uvrg_xxxx0));
 854
 855   gvdup_l_u16(tmp_lo, gvhi(v0), 0);     // xxxx0 = { xx0, xx0 }
 856   gvset_hi(uvrg_xxxx0, tmp_lo);
 857
 858   u32 x1_x2 = (u16)b->x | (c->x << 16); // x1_x2 = { x1, x2 }
 859   u32 x0_x1 = (u16)a->x | (b->x << 16); // x0_x1 = { x0, x1 }
 860
 861   gvdup_l_u16(tmp_lo, gvhi(v1), 1);     // yyyy1 = { yy1, yy1 }
 862   gvcombine_u16(y1_ab, tmp_lo, gvlo(uvrg_xxxx1));
 863
 864   gvdup_l_u16(tmp_lo, gvhi(v1), 0);     // xxxx1 = { xx1, xx1 }
 865   gvset_hi(uvrg_xxxx1, tmp_lo);
 866
 867   gvdup_l_u16(tmp_lo, gvhi(v2), 1);     // yyyy2 = { yy2, yy2 }
 868   gvcombine_u16(y2_ab, tmp_lo, gvlo(uvrg_xxxx2));
 869
 870   gvdup_l_u16(tmp_lo, gvhi(v2), 0);     // xxxx2 = { xx2, xx2 }
 871   gvset_hi(uvrg_xxxx2, tmp_lo);
 872
 873   u32 y0_y1 = (u16)a->y | (b->y << 16); // y0_y1 = { y0, y1 }
 874   u32 y1_y2 = (u16)b->y | (c->y << 16); // y1_y2 = { y1, y2 }
 875
 876   gvsubq_u16(d0_ab, uvrg_xxxx1, uvrg_xxxx0);
 877
 878   u32 b1_b2 = b->b | (c->b << 16);      // b1_b2 = { b1, b2 }
 879
 880   gvsubq_u16(d2_ab, uvrg_xxxx2, uvrg_xxxx1);
 881
 882   gvsubq_u16(d1_ab, y2_ab, y1_ab);
 883
 884   u32 b0_b1 = a->b | (b->b << 16);      // b0_b1 = { b0, b1 }
 885
 886   u32 dx, dy, db;
 887   gssub16(dx, x1_x2, x0_x1);            // dx = { x1 - x0, x2 - x1 }
 888   gssub16(dy, y1_y2, y0_y1);            // dy = { y1 - y0, y2 - y1 }
 889   gssub16(db, b1_b2, b0_b1);            // db = { b1 - b0, b2 - b1 }
 890
 891   u32 ga_by, ga_bx;
 892   gvsubq_u16(d3_ab, y1_ab, y0_ab);
 893   gsmusdx(ga_by, dx, db);               // ga_by = ((x1 - x0) * (b2 - b1)) -
 894                                         //         ((x2 - X1) * (b1 - b0))
 895   gvmull_s16(ga_uvrg_x, gvlo(d0_ab), gvlo(d1_ab));
 896   gsmusdx(ga_bx, db, dy);               // ga_bx = ((b1 - b0) * (y2 - y1)) -
 897                                         //         ((b2 - b1) * (y1 - y0))
 898   gvmlsl_s16(ga_uvrg_x, gvlo(d2_ab), gvlo(d3_ab));
 899   u32 gs_bx = (s32)ga_bx >> 31;         // movs
 900
 901   gvmull_s16(ga_uvrg_y, gvhi(d0_ab), gvhi(d1_ab));
 902   if ((s32)gs_bx < 0) ga_bx = -ga_bx;   // rsbmi
 903
 904   gvmlsl_s16(ga_uvrg_y, gvhi(d2_ab), gvhi(d3_ab));
 905   u32 gs_by = (s32)ga_by >> 31;         // movs
 906
 907   gvhreg d0;
 908   gvshr_n_u64(d0, d30.v, 22);           // note: on "d30 >> 22" gcc generates junk code
 909
 910   gvdupq_n_u32(uvrgb_phase, psx_gpu->uvrgb_phase);
 911   u32 b_base = psx_gpu->uvrgb_phase + (a->b << 16);
 912
 913   if ((s32)gs_by < 0) ga_by = -ga_by;   // rsbmi
 914   gvreg gs_uvrg_x, gs_uvrg_y;
 915   gs_uvrg_x.s32 = ga_uvrg_x.s32 < 0;    // gs_uvrg_x = ga_uvrg_x < 0
 916   gs_uvrg_y.s32 = ga_uvrg_y.s32 < 0;    // gs_uvrg_y = ga_uvrg_y < 0
 917
 918   gvdupq_n_u32(w_mask, -psx_gpu->triangle_winding); // w_mask = { -w, -w, -w, -w }
 919   shift -= 62 - 12;                     // shift -= (62 - FIXED_BITS)
 920
 921   gvreg uvrg_base;
 922   gvshll_n_u16(uvrg_base, gvlo(uvrg_xxxx0), 16); // uvrg_base = uvrg0 << 16
 923
 924   gvaddq_u32(uvrg_base, uvrg_base, uvrgb_phase);
 925   gvabsq_s32(ga_uvrg_x, ga_uvrg_x);     // ga_uvrg_x = abs(ga_uvrg_x)
 926
 927   u32 area_r_s = d0.u32[0];             // area_r_s = triangle_reciprocal
 928   gvabsq_s32(ga_uvrg_y, ga_uvrg_y);     // ga_uvrg_y = abs(ga_uvrg_y)
 929
 930   gvmull_l_u32(gw_rg_x, gvhi(ga_uvrg_x), d0, 0);
 931   gvmull_l_u32(ga_uvrg_x, gvlo(ga_uvrg_x), d0, 0);
 932   gvmull_l_u32(gw_rg_y, gvhi(ga_uvrg_y), d0, 0);
 933   gvmull_l_u32(ga_uvrg_y, gvlo(ga_uvrg_y), d0, 0);
 934
 935 #if defined(__ARM_NEON) || defined(__ARM_NEON__)
 936   gvdupq_n_s64(r_shift, shift);         // r_shift = { shift, shift }
 937   gvshlq_u64(gw_rg_x, gw_rg_x, r_shift);
 938   gvshlq_u64(ga_uvrg_x, ga_uvrg_x, r_shift);
 939   gvshlq_u64(gw_rg_y, gw_rg_y, r_shift);
 940   gvshlq_u64(ga_uvrg_y, ga_uvrg_y, r_shift);
 941 #elif defined(__SSE2__)
 942   r_shift.m   = _mm_cvtsi32_si128(-shift);
 943   gw_rg_x.m   = _mm_srl_epi64(gw_rg_x.m, r_shift.m);
 944   ga_uvrg_x.m = _mm_srl_epi64(ga_uvrg_x.m, r_shift.m);
 945   gw_rg_y.m   = _mm_srl_epi64(gw_rg_y.m, r_shift.m);
 946   ga_uvrg_y.m = _mm_srl_epi64(ga_uvrg_y.m, r_shift.m);
 947 #else
 948   gvdupq_n_s64(r_shift, -shift);        // r_shift = { shift, shift }
 949   gvshrq_u64(gw_rg_x, gw_rg_x, r_shift);
 950   gvshrq_u64(ga_uvrg_x, ga_uvrg_x, r_shift);
 951   gvshrq_u64(gw_rg_y, gw_rg_y, r_shift);
 952   gvshrq_u64(ga_uvrg_y, ga_uvrg_y, r_shift);
 953 #endif
 954
 955   gveorq(gs_uvrg_x, gs_uvrg_x, w_mask);
 956   gvmovn_u64(tmp_lo, ga_uvrg_x);
 957
 958   gveorq(gs_uvrg_y, gs_uvrg_y, w_mask);
 959   gvmovn_u64(tmp_hi, gw_rg_x);
 960
 961   gvcombine_u32(ga_uvrg_x, tmp_lo, tmp_hi);
 962
 963   gveorq(ga_uvrg_x, ga_uvrg_x, gs_uvrg_x);
 964   gvmovn_u64(tmp_lo, ga_uvrg_y);
 965
 966   gvsubq_u32(ga_uvrg_x, ga_uvrg_x, gs_uvrg_x);
 967   gvmovn_u64(tmp_hi, gw_rg_y);
 968
 969   gvcombine_u32(ga_uvrg_y, tmp_lo, tmp_hi);
 970
 971   gveorq(ga_uvrg_y, ga_uvrg_y, gs_uvrg_y);
 972   ga_bx = ga_bx << 13;
 973
 974   gvsubq_u32(ga_uvrg_y, ga_uvrg_y, gs_uvrg_y);
 975   ga_by = ga_by << 13;
 976
 977   u32 gw_bx_h, gw_by_h;
 978   gw_bx_h = (u64)ga_bx * area_r_s >> 32;
 979
 980   gvshlq_n_u32(ga_uvrg_x, ga_uvrg_x, 4);
 981   gvshlq_n_u32(ga_uvrg_y, ga_uvrg_y, 4);
 982
 983   gw_by_h = (u64)ga_by * area_r_s >> 32;
 984   gvdup_n_u32(tmp_lo, a->x);
 985   gvmlsq_l_s32(uvrg_base, ga_uvrg_x, tmp_lo, 0);
 986
 987   gs_bx = gs_bx ^ -psx_gpu->triangle_winding;
 988   gvaddq_u32(uvrg_dx2, ga_uvrg_x, ga_uvrg_x);
 989
 990   gs_by = gs_by ^ -psx_gpu->triangle_winding;
 991
 992   u32 r11 = -shift;                         // r11 = negative shift for scalar lsr
 993   u32 *store_a = psx_gpu->uvrg.e;
 994   r11 = r11 - (32 - 13);
 995   u32 *store_b = store_a + 16 / sizeof(u32);
 996
 997   gvaddq_u32(uvrg_dx3, uvrg_dx2, ga_uvrg_x);
 998   gvst1q_inc_u32(uvrg_base, store_a, 32);
 999
1000   gvst1q_inc_u32(ga_uvrg_x, store_b, 32);
1001   u32 g_bx = (u32)gw_bx_h >> r11;
1002
1003   gvst1q_inc_u32(ga_uvrg_y, store_a, 32);
1004   u32 g_by = (u32)gw_by_h >> r11;
1005
1006   gvdup_n_u32(zero, 0);
1007
1008   gvst4_4_inc_u32(zero, gvlo(ga_uvrg_x), gvlo(uvrg_dx2), gvlo(uvrg_dx3), store_b, 32);
1009   g_bx = g_bx ^ gs_bx;
1010
1011   gvst4_4_inc_u32(zero, gvhi(ga_uvrg_x), gvhi(uvrg_dx2), gvhi(uvrg_dx3), store_b, 32);
1012   g_bx = g_bx - gs_bx;
1013
1014   g_bx = g_bx << 4;
1015   g_by = g_by ^ gs_by;
1016
1017   b_base -= g_bx * a->x;
1018   g_by = g_by - gs_by;
1019
1020   g_by = g_by << 4;
1021
1022   u32 g_bx2 = g_bx + g_bx;
1023   u32 g_bx3 = g_bx + g_bx2;
1024
1025   // 112
1026   store_b[0] = 0;
1027   store_b[1] = g_bx;
1028   store_b[2] = g_bx2;
1029   store_b[3] = g_bx3;
1030   store_b[4] = b_base;
1031   store_b[5] = g_by; // 132
1032 }
1033
1034 #define setup_spans_debug_check(span_edge_data_element)                        \
1035
1036 #define setup_spans_prologue_alternate_yes()                                   \
1037   vec_2x64s alternate_x;                                                       \
1038   vec_2x64s alternate_dx_dy;                                                   \
1039   vec_4x32s alternate_x_32;                                                    \
1040   vec_4x16u alternate_x_16;                                                    \
1041                                                                                \
1042   vec_4x16u alternate_select;                                                  \
1043   vec_4x16s y_mid_point;                                                       \
1044                                                                                \
1045   s32 y_b = v_b->y;                                                            \
1046   s64 edge_alt;                                                                \
1047   s32 edge_dx_dy_alt;                                                          \
1048   u32 edge_shift_alt                                                           \
1049
1050 #define setup_spans_prologue_alternate_no()                                    \
1051
1052 #define setup_spans_prologue(alternate_active)                                 \
1053   edge_data_struct *span_edge_data;                                            \
1054   vec_4x32u *span_uvrg_offset;                                                 \
1055   u32 *span_b_offset;                                                          \
1056                                                                                \
1057   s32 clip;                                                                    \
1058   vec_4x32u v_clip;                                                            \
1059                                                                                \
1060   vec_2x64s edges_xy;                                                          \
1061   vec_2x32s edges_dx_dy;                                                       \
1062   vec_2x32u edge_shifts;                                                       \
1063                                                                                \
1064   vec_2x64s left_x, right_x;                                                   \
1065   vec_2x64s left_dx_dy, right_dx_dy;                                           \
1066   vec_4x32s left_x_32, right_x_32;                                             \
1067   vec_2x32s left_x_32_lo, right_x_32_lo;                                       \
1068   vec_2x32s left_x_32_hi, right_x_32_hi;                                       \
1069   vec_4x16s left_right_x_16_lo, left_right_x_16_hi;                            \
1070   vec_4x16s y_x4;                                                              \
1071   vec_8x16s left_edge;                                                         \
1072   vec_8x16s right_edge;                                                        \
1073   vec_4x16u span_shift;                                                        \
1074                                                                                \
1075   vec_2x32u c_0x01;                                                            \
1076   vec_4x16u c_0x04;                                                            \
1077   vec_4x16u c_0xFFFE;                                                          \
1078   vec_4x16u c_0x07;                                                            \
1079                                                                                \
1080   vec_2x32s x_starts;                                                          \
1081   vec_2x32s x_ends;                                                            \
1082                                                                                \
1083   s32 x_a = v_a->x;                                                            \
1084   s32 x_b = v_b->x;                                                            \
1085   s32 x_c = v_c->x;                                                            \
1086   s32 y_a = v_a->y;                                                            \
1087   s32 y_c = v_c->y;                                                            \
1088                                                                                \
1089   vec_4x32u uvrg;                                                              \
1090   vec_4x32u uvrg_dy;                                                           \
1091   u32 b = psx_gpu->b;                                                          \
1092   u32 b_dy = psx_gpu->b_dy;                                                    \
1093   const u32 *reciprocal_table = psx_gpu->reciprocal_table_ptr;                 \
1094                                                                                \
1095   gvld1q_u32(uvrg, psx_gpu->uvrg.e);                                           \
1096   gvld1q_u32(uvrg_dy, psx_gpu->uvrg_dy.e);                                     \
1097   gvdup_n_u32(c_0x01, 0x01);                                                   \
1098   setup_spans_prologue_alternate_##alternate_active()                          \
1099
1100 #define setup_spans_prologue_b()                                               \
1101   span_edge_data = psx_gpu->span_edge_data;                                    \
1102   span_uvrg_offset = (vec_4x32u *)psx_gpu->span_uvrg_offset;                   \
1103   span_b_offset = psx_gpu->span_b_offset;                                      \
1104                                                                                \
1105   vec_8x16u c_0x0001;                                                          \
1106   vec_4x16u c_max_blocks_per_row;                                              \
1107                                                                                \
1108   gvdupq_n_u16(c_0x0001, 0x0001);                                              \
1109   gvdupq_n_u16(left_edge, psx_gpu->viewport_start_x);                          \
1110   gvdupq_n_u16(right_edge, psx_gpu->viewport_end_x);                           \
1111   gvaddq_u16(right_edge, right_edge, c_0x0001);                                \
1112   gvdup_n_u16(c_0x04, 0x04);                                                   \
1113   gvdup_n_u16(c_0x07, 0x07);                                                   \
1114   gvdup_n_u16(c_0xFFFE, 0xFFFE);                                               \
1115   gvdup_n_u16(c_max_blocks_per_row, MAX_BLOCKS_PER_ROW);                       \
1116
1117 #if defined(__ARM_NEON) || defined(__ARM_NEON__)
1118 // better encoding, remaining bits are unused anyway
1119 #define mask_edge_shifts(edge_shifts)                                          \
1120   gvbic_n_u16(edge_shifts, 0xE0)
1121 #else
1122 #define mask_edge_shifts(edge_shifts)                                          \
1123   gvand_n_u32(edge_shifts, 0x1F)
1124 #endif
1125
1126 #define compute_edge_delta_x2()                                                \
1127 {                                                                              \
1128   vec_2x32s heights;                                                           \
1129   vec_2x32s height_reciprocals;                                                \
1130   vec_2x32s heights_b;                                                         \
1131   vec_2x32u widths;                                                            \
1132                                                                                \
1133   u32 edge_shift = reciprocal_table[height];                                   \
1134                                                                                \
1135   gvdup_n_u32(heights, height);                                                \
1136   gvsub_u32(widths, x_ends, x_starts);                                         \
1137                                                                                \
1138   gvdup_n_u32(edge_shifts, edge_shift);                                        \
1139   gvsub_u32(heights_b, heights, c_0x01);                                       \
1140   gvshr_n_u32(height_reciprocals, edge_shifts, 10);                            \
1141                                                                                \
1142   gvmla_s32(heights_b, x_starts, heights);                                     \
1143   mask_edge_shifts(edge_shifts);                                               \
1144   gvmul_s32(edges_dx_dy, widths, height_reciprocals);                          \
1145   gvmull_s32(edges_xy, heights_b, height_reciprocals);                         \
1146 }                                                                              \
1147
1148 #define compute_edge_delta_x3(start_c, height_a, height_b)                     \
1149 {                                                                              \
1150   vec_2x32s heights;                                                           \
1151   vec_2x32s height_reciprocals;                                                \
1152   vec_2x32s heights_b;                                                         \
1153   vec_2x32u widths;                                                            \
1154                                                                                \
1155   u32 width_alt;                                                               \
1156   s32 height_b_alt;                                                            \
1157   u32 height_reciprocal_alt;                                                   \
1158                                                                                \
1159   gvcreate_u32(heights, height_a, height_b);                                   \
1160   gvcreate_u32(edge_shifts, reciprocal_table[height_a], reciprocal_table[height_b]); \
1161                                                                                \
1162   edge_shift_alt = reciprocal_table[height_minor_b];                           \
1163                                                                                \
1164   gvsub_u32(widths, x_ends, x_starts);                                         \
1165   width_alt = x_c - start_c;                                                   \
1166                                                                                \
1167   gvshr_n_u32(height_reciprocals, edge_shifts, 10);                            \
1168   height_reciprocal_alt = edge_shift_alt >> 10;                                \
1169                                                                                \
1170   mask_edge_shifts(edge_shifts);                                               \
1171   edge_shift_alt &= 0x1F;                                                      \
1172                                                                                \
1173   gvsub_u32(heights_b, heights, c_0x01);                                       \
1174   height_b_alt = height_minor_b - 1;                                           \
1175                                                                                \
1176   gvmla_s32(heights_b, x_starts, heights);                                     \
1177   height_b_alt += height_minor_b * start_c;                                    \
1178                                                                                \
1179   gvmull_s32(edges_xy, heights_b, height_reciprocals);                         \
1180   edge_alt = (s64)height_b_alt * height_reciprocal_alt;                        \
1181                                                                                \
1182   gvmul_s32(edges_dx_dy, widths, height_reciprocals);                          \
1183   edge_dx_dy_alt = width_alt * height_reciprocal_alt;                          \
1184 }                                                                              \
1185
1186
1187 #define setup_spans_adjust_y_up()                                              \
1188   gvsub_u32(y_x4, y_x4, c_0x04)                                                \
1189
1190 #define setup_spans_adjust_y_down()                                            \
1191   gvadd_u32(y_x4, y_x4, c_0x04)                                                \
1192
1193 #define setup_spans_adjust_interpolants_up()                                   \
1194   gvsubq_u32(uvrg, uvrg, uvrg_dy);                                             \
1195   b -= b_dy                                                                    \
1196
1197 #define setup_spans_adjust_interpolants_down()                                 \
1198   gvaddq_u32(uvrg, uvrg, uvrg_dy);                                             \
1199   b += b_dy                                                                    \
1200
1201
1202 #define setup_spans_clip_interpolants_increment()                              \
1203   gvmlaq_s32(uvrg, uvrg_dy, v_clip);                                           \
1204   b += b_dy * clip                                                             \
1205
1206 #define setup_spans_clip_interpolants_decrement()                              \
1207   gvmlsq_s32(uvrg, uvrg_dy, v_clip);                                           \
1208   b -= b_dy * clip                                                             \
1209
1210 #define setup_spans_clip_alternate_yes()                                       \
1211   edge_alt += edge_dx_dy_alt * (s64)(clip)                                     \
1212
1213 #define setup_spans_clip_alternate_no()                                        \
1214
1215 #define setup_spans_clip(direction, alternate_active)                          \
1216 {                                                                              \
1217   gvdupq_n_u32(v_clip, clip);                                                  \
1218   gvmlal_s32(edges_xy, edges_dx_dy, gvlo(v_clip));                             \
1219   setup_spans_clip_alternate_##alternate_active();                             \
1220   setup_spans_clip_interpolants_##direction();                                 \
1221 }                                                                              \
1222
1223
1224 #define setup_spans_adjust_edges_alternate_no(left_half, right_half)           \
1225 {                                                                              \
1226   vec_2x64s edge_shifts_64;                                                    \
1227   vec_2x64s edges_dx_dy_64;                                                    \
1228   vec_1x64s left_x_hi, right_x_hi;                                             \
1229                                                                                \
1230   gvmovl_s32(edge_shifts_64, edge_shifts);                                     \
1231   gvshlq_u64(edges_xy, edges_xy, edge_shifts_64);                              \
1232                                                                                \
1233   gvmovl_s32(edges_dx_dy_64, edges_dx_dy);                                     \
1234   gvshlq_u64(edges_dx_dy_64, edges_dx_dy_64, edge_shifts_64);                  \
1235                                                                                \
1236   gvdupq_l_s64(left_x, gv##left_half(edges_xy), 0);                            \
1237   gvdupq_l_s64(right_x, gv##right_half(edges_xy), 0);                          \
1238                                                                                \
1239   gvdupq_l_s64(left_dx_dy, gv##left_half(edges_dx_dy_64), 0);                  \
1240   gvdupq_l_s64(right_dx_dy, gv##right_half(edges_dx_dy_64), 0);                \
1241                                                                                \
1242   gvadd_s64(left_x_hi, gvlo(left_x), gvlo(left_dx_dy));                        \
1243   gvadd_s64(right_x_hi, gvlo(right_x), gvlo(right_dx_dy));                     \
1244                                                                                \
1245   gvset_hi(left_x, left_x_hi);                                                 \
1246   gvset_hi(right_x, right_x_hi);                                               \
1247                                                                                \
1248   gvaddq_s64(left_dx_dy, left_dx_dy, left_dx_dy);                              \
1249   gvaddq_s64(right_dx_dy, right_dx_dy, right_dx_dy);                           \
1250 }                                                                              \
1251
1252 #define setup_spans_adjust_edges_alternate_yes(left_half, right_half)          \
1253 {                                                                              \
1254   setup_spans_adjust_edges_alternate_no(left_half, right_half);                \
1255   s64 edge_dx_dy_alt_64;                                                       \
1256   vec_1x64s alternate_x_hi;                                                    \
1257                                                                                \
1258   gvdup_n_u16(y_mid_point, y_b);                                               \
1259                                                                                \
1260   edge_alt <<= edge_shift_alt;                                                 \
1261   edge_dx_dy_alt_64 = (s64)edge_dx_dy_alt << edge_shift_alt;                   \
1262                                                                                \
1263   gvdupq_n_s64(alternate_x, edge_alt);                                         \
1264   gvdupq_n_s64(alternate_dx_dy, edge_dx_dy_alt_64);                            \
1265                                                                                \
1266   gvadd_s64(alternate_x_hi, gvlo(alternate_x), gvlo(alternate_dx_dy));         \
1267   gvaddq_s64(alternate_dx_dy, alternate_dx_dy, alternate_dx_dy);               \
1268   gvset_hi(alternate_x, alternate_x_hi);                                       \
1269 }                                                                              \
1270
1271
1272 #define setup_spans_y_select_up()                                              \
1273   gvclt_s16(alternate_select, y_x4, y_mid_point)                               \
1274
1275 #define setup_spans_y_select_down()                                            \
1276   gvcgt_s16(alternate_select, y_x4, y_mid_point)                               \
1277
1278 #define setup_spans_y_select_alternate_yes(direction)                          \
1279   setup_spans_y_select_##direction()                                           \
1280
1281 #define setup_spans_y_select_alternate_no(direction)                           \
1282
1283 #define setup_spans_alternate_select_left()                                    \
1284   gvbit(left_right_x_16_lo, alternate_x_16, alternate_select);                 \
1285
1286 #define setup_spans_alternate_select_right()                                   \
1287   gvbit(left_right_x_16_hi, alternate_x_16, alternate_select);                 \
1288
1289 #define setup_spans_alternate_select_none()                                    \
1290
1291 #define setup_spans_increment_alternate_yes()                                  \
1292 {                                                                              \
1293   vec_2x32s alternate_x_32_lo, alternate_x_32_hi;                              \
1294   gvmovn_top_u64(alternate_x_32_lo, alternate_x);                              \
1295   gvaddq_s64(alternate_x, alternate_x, alternate_dx_dy);                       \
1296   gvmovn_top_u64(alternate_x_32_hi, alternate_x);                              \
1297   gvaddq_s64(alternate_x, alternate_x, alternate_dx_dy);                       \
1298   gvcombine_u32(alternate_x_32, alternate_x_32_lo, alternate_x_32_hi);         \
1299   gvmovn_u32(alternate_x_16, alternate_x_32);                                  \
1300 }                                                                              \
1301
1302 #define setup_spans_increment_alternate_no()                                   \
1303
1304 #if defined(__SSE2__) && !(defined(__AVX512BW__) && defined(__AVX512VL__))
1305 #define setup_spans_make_span_shift(span_shift) {                              \
1306   gvreg tab1_ = { .u8 = { 0xfe, 0xfc, 0xf8, 0xf0, 0xe0, 0xc0, 0x80, 0x00 } };  \
1307   gvtbl2_u8(span_shift, tab1_, span_shift);                                    \
1308   gvorr_n_u16(span_shift, 0xff00);                                             \
1309   (void)c_0xFFFE;                                                              \
1310 }
1311 #else
1312 #define setup_spans_make_span_shift(span_shift)                                \
1313   gvshl_u16(span_shift, c_0xFFFE, span_shift)
1314 #endif
1315
1316 #define setup_spans_set_x4(alternate, direction, alternate_active)             \
1317 {                                                                              \
1318   gvst1q_pi_u32(uvrg, span_uvrg_offset);                                       \
1319   *span_b_offset++ = b;                                                        \
1320   setup_spans_adjust_interpolants_##direction();                               \
1321                                                                                \
1322   gvst1q_pi_u32(uvrg, span_uvrg_offset);                                       \
1323   *span_b_offset++ = b;                                                        \
1324   setup_spans_adjust_interpolants_##direction();                               \
1325                                                                                \
1326   gvst1q_pi_u32(uvrg, span_uvrg_offset);                                       \
1327   *span_b_offset++ = b;                                                        \
1328   setup_spans_adjust_interpolants_##direction();                               \
1329                                                                                \
1330   gvst1q_pi_u32(uvrg, span_uvrg_offset);                                       \
1331   *span_b_offset++ = b;                                                        \
1332   setup_spans_adjust_interpolants_##direction();                               \
1333                                                                                \
1334   gvmovn_top_u64(left_x_32_lo, left_x);                                        \
1335   gvmovn_top_u64(right_x_32_lo, right_x);                                      \
1336                                                                                \
1337   gvaddq_s64(left_x, left_x, left_dx_dy);                                      \
1338   gvaddq_s64(right_x, right_x, right_dx_dy);                                   \
1339                                                                                \
1340   gvmovn_top_u64(left_x_32_hi, left_x);                                        \
1341   gvmovn_top_u64(right_x_32_hi, right_x);                                      \
1342                                                                                \
1343   gvaddq_s64(left_x, left_x, left_dx_dy);                                      \
1344   gvaddq_s64(right_x, right_x, right_dx_dy);                                   \
1345                                                                                \
1346   gvcombine_s64(left_x_32, left_x_32_lo, left_x_32_hi);                        \
1347   gvcombine_s64(right_x_32, right_x_32_lo, right_x_32_hi);                     \
1348                                                                                \
1349   gvmovn_u32(left_right_x_16_lo, left_x_32);                                   \
1350   gvmovn_u32(left_right_x_16_hi, right_x_32);                                  \
1351                                                                                \
1352   setup_spans_increment_alternate_##alternate_active();                        \
1353   setup_spans_y_select_alternate_##alternate_active(direction);                \
1354   setup_spans_alternate_select_##alternate();                                  \
1355                                                                                \
1356   gvmax_s16(left_right_x_16_lo, left_right_x_16_lo, gvlo(left_edge));          \
1357   gvmax_s16(left_right_x_16_hi, left_right_x_16_hi, gvhi(left_edge));          \
1358   gvmin_s16(left_right_x_16_lo, left_right_x_16_lo, gvlo(right_edge));         \
1359   gvmin_s16(left_right_x_16_hi, left_right_x_16_hi, gvhi(right_edge));         \
1360                                                                                \
1361   gvsub_u16(left_right_x_16_hi, left_right_x_16_hi, left_right_x_16_lo);       \
1362   gvadd_u16(left_right_x_16_hi, left_right_x_16_hi, c_0x07);                   \
1363   gvand(span_shift, left_right_x_16_hi, c_0x07);                               \
1364   setup_spans_make_span_shift(span_shift);                                     \
1365   gvshr_n_u16(left_right_x_16_hi, left_right_x_16_hi, 3);                      \
1366   gvmin_u16(left_right_x_16_hi, left_right_x_16_hi, c_max_blocks_per_row);     \
1367                                                                                \
1368   gvst4_pi_u16(left_right_x_16_lo, left_right_x_16_hi, span_shift, y_x4,       \
1369     span_edge_data);                                                           \
1370                                                                                \
1371   setup_spans_adjust_y_##direction();                                          \
1372 }                                                                              \
1373
1374
1375 #define setup_spans_alternate_adjust_yes()                                     \
1376   edge_alt -= edge_dx_dy_alt * (s64)height_minor_a                             \
1377
1378 #define setup_spans_alternate_adjust_no()                                      \
1379
1380
1381 #define setup_spans_down(left_half, right_half, alternate, alternate_active)   \
1382   setup_spans_alternate_adjust_##alternate_active();                           \
1383   if(y_c > psx_gpu->viewport_end_y)                                            \
1384     height -= y_c - psx_gpu->viewport_end_y - 1;                               \
1385                                                                                \
1386   clip = psx_gpu->viewport_start_y - y_a;                                      \
1387   if(clip > 0)                                                                 \
1388   {                                                                            \
1389     height -= clip;                                                            \
1390     y_a += clip;                                                               \
1391     setup_spans_clip(increment, alternate_active);                             \
1392   }                                                                            \
1393                                                                                \
1394   setup_spans_prologue_b();                                                    \
1395                                                                                \
1396   if (height > 512)                                                            \
1397     height = 512;                                                              \
1398   if (height > 0)                                                              \
1399   {                                                                            \
1400     u64 y_x4_ = ((u64)(y_a + 3) << 48) | ((u64)(u16)(y_a + 2) << 32)           \
1401               | (u32)((y_a + 1) << 16) | (u16)y_a;                             \
1402     gvcreate_u64(y_x4, y_x4_);                                                 \
1403     setup_spans_adjust_edges_alternate_##alternate_active(left_half, right_half); \
1404                                                                                \
1405     psx_gpu->num_spans = height;                                               \
1406     do                                                                         \
1407     {                                                                          \
1408       setup_spans_set_x4(alternate, down, alternate_active);                   \
1409       height -= 4;                                                             \
1410     } while(height > 0);                                                       \
1411     if (psx_gpu->hacks_active & (AHACK_TEXTURE_ADJ_U | AHACK_TEXTURE_ADJ_V))   \
1412     {                                                                          \
1413       vec_2x32u tmp;                                                           \
1414       gvld1_u64(tmp, &span_uvrg_offset[height - 2]);                           \
1415       gvst1_u64(tmp, &span_uvrg_offset[height - 1]);                           \
1416     }                                                                          \
1417   }                                                                            \
1418
1419
1420 #define setup_spans_alternate_pre_increment_yes()                              \
1421   edge_alt += edge_dx_dy_alt                                                   \
1422
1423 #define setup_spans_alternate_pre_increment_no()                               \
1424
1425 #define setup_spans_up_decrement_height_yes()                                  \
1426   height--                                                                     \
1427
1428 #define setup_spans_up_decrement_height_no()                                   \
1429   {}                                                                           \
1430
1431 #define setup_spans_up(left_half, right_half, alternate, alternate_active)     \
1432   setup_spans_alternate_adjust_##alternate_active();                           \
1433   y_a--;                                                                       \
1434                                                                                \
1435   if(y_c < psx_gpu->viewport_start_y)                                          \
1436     height -= psx_gpu->viewport_start_y - y_c;                                 \
1437   else                                                                         \
1438     setup_spans_up_decrement_height_##alternate_active();                      \
1439                                                                                \
1440   clip = y_a - psx_gpu->viewport_end_y;                                        \
1441   if(clip > 0)                                                                 \
1442   {                                                                            \
1443     height -= clip;                                                            \
1444     y_a -= clip;                                                               \
1445     setup_spans_clip(decrement, alternate_active);                             \
1446   }                                                                            \
1447                                                                                \
1448   setup_spans_prologue_b();                                                    \
1449                                                                                \
1450   if (height > 512)                                                            \
1451     height = 512;                                                              \
1452   if (height > 0)                                                              \
1453   {                                                                            \
1454     u64 y_x4_ = ((u64)(y_a - 3) << 48) | ((u64)(u16)(y_a - 2) << 32)           \
1455               | (u32)((y_a - 1) << 16) | (u16)y_a;                             \
1456     gvcreate_u64(y_x4, y_x4_);                                                 \
1457     gvaddw_s32(edges_xy, edges_xy, edges_dx_dy);                               \
1458     setup_spans_alternate_pre_increment_##alternate_active();                  \
1459     setup_spans_adjust_edges_alternate_##alternate_active(left_half, right_half); \
1460     setup_spans_adjust_interpolants_up();                                      \
1461                                                                                \
1462     psx_gpu->num_spans = height;                                               \
1463     while(height > 0)                                                          \
1464     {                                                                          \
1465       setup_spans_set_x4(alternate, up, alternate_active);                     \
1466       height -= 4;                                                             \
1467     }                                                                          \
1468     if (psx_gpu->hacks_active & AHACK_TEXTURE_ADJ_V)                           \
1469     {                                                                          \
1470       vec_2x32u tmp;                                                           \
1471       gvld1_u64(tmp, &psx_gpu->span_uvrg_offset[1]);                           \
1472       gvst1_u64(tmp, &psx_gpu->span_uvrg_offset[0]);                           \
1473     }                                                                          \
1474   }                                                                            \
1475
1476 #define half_left  lo
1477 #define half_right hi
1478
1479 #define setup_spans_up_up(minor, major)                                        \
1480   setup_spans_prologue(yes);                                                   \
1481   s32 height_minor_a = y_a - y_b;                                              \
1482   s32 height_minor_b = y_b - y_c;                                              \
1483   s32 height = y_a - y_c;                                                      \
1484                                                                                \
1485   gvdup_n_u32(x_starts, x_a);                                                  \
1486   gvcreate_u32(x_ends, x_c, x_b);                                              \
1487                                                                                \
1488   compute_edge_delta_x3(x_b, height, height_minor_a);                          \
1489   setup_spans_up(half_##major, half_##minor, minor, yes)                       \
1490
1491 void setup_spans_up_left(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1492  vertex_struct *v_b, vertex_struct *v_c)
1493 {
1494 #if 0
1495   setup_spans_up_left_(psx_gpu, v_a, v_b, v_c);
1496   return;
1497 #endif
1498   setup_spans_up_up(left, right)
1499 }
1500
1501 void setup_spans_up_right(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1502  vertex_struct *v_b, vertex_struct *v_c)
1503 {
1504 #if 0
1505   setup_spans_up_right_(psx_gpu, v_a, v_b, v_c);
1506   return;
1507 #endif
1508   setup_spans_up_up(right, left)
1509 }
1510
1511 #define setup_spans_down_down(minor, major)                                    \
1512   setup_spans_prologue(yes);                                                   \
1513   s32 height_minor_a = y_b - y_a;                                              \
1514   s32 height_minor_b = y_c - y_b;                                              \
1515   s32 height = y_c - y_a;                                                      \
1516                                                                                \
1517   gvdup_n_u32(x_starts, x_a);                                                  \
1518   gvcreate_u32(x_ends, x_c, x_b);                                              \
1519                                                                                \
1520   compute_edge_delta_x3(x_b, height, height_minor_a);                          \
1521   setup_spans_down(half_##major, half_##minor, minor, yes)                     \
1522
1523 void setup_spans_down_left(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1524  vertex_struct *v_b, vertex_struct *v_c)
1525 {
1526 #if 0
1527   setup_spans_down_left_(psx_gpu, v_a, v_b, v_c);
1528   return;
1529 #endif
1530   setup_spans_down_down(left, right)
1531 }
1532
1533 void setup_spans_down_right(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1534  vertex_struct *v_b, vertex_struct *v_c)
1535 {
1536 #if 0
1537   setup_spans_down_right_(psx_gpu, v_a, v_b, v_c);
1538   return;
1539 #endif
1540   setup_spans_down_down(right, left)
1541 }
1542
1543 #define setup_spans_up_flat()                                                  \
1544   s32 height = y_a - y_c;                                                      \
1545                                                                                \
1546   compute_edge_delta_x2();                                                     \
1547   setup_spans_up(half_left, half_right, none, no)                              \
1548
1549 void setup_spans_up_a(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1550  vertex_struct *v_b, vertex_struct *v_c)
1551 {
1552 #if 0
1553   setup_spans_up_a_(psx_gpu, v_a, v_b, v_c);
1554   return;
1555 #endif
1556   setup_spans_prologue(no);
1557
1558   gvcreate_u32(x_starts, x_a, x_b);
1559   gvdup_n_u32(x_ends, x_c);
1560
1561   setup_spans_up_flat()
1562 }
1563
1564 void setup_spans_up_b(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1565  vertex_struct *v_b, vertex_struct *v_c)
1566 {
1567 #if 0
1568   setup_spans_up_b_(psx_gpu, v_a, v_b, v_c);
1569   return;
1570 #endif
1571   setup_spans_prologue(no);
1572
1573   gvdup_n_u32(x_starts, x_a);
1574   gvcreate_u32(x_ends, x_b, x_c);
1575
1576   setup_spans_up_flat()
1577 }
1578
1579 #define setup_spans_down_flat()                                                \
1580   s32 height = y_c - y_a;                                                      \
1581                                                                                \
1582   compute_edge_delta_x2();                                                     \
1583   setup_spans_down(half_left, half_right, none, no)                            \
1584
1585 void setup_spans_down_a(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1586  vertex_struct *v_b, vertex_struct *v_c)
1587 {
1588 #if 0
1589   setup_spans_down_a_(psx_gpu, v_a, v_b, v_c);
1590   return;
1591 #endif
1592   setup_spans_prologue(no);
1593
1594   gvcreate_u32(x_starts, x_a, x_b);
1595   gvdup_n_u32(x_ends, x_c);
1596
1597   setup_spans_down_flat()
1598 }
1599
1600 void setup_spans_down_b(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1601  vertex_struct *v_b, vertex_struct *v_c)
1602 {
1603 #if 0
1604   setup_spans_down_b_(psx_gpu, v_a, v_b, v_c);
1605   return;
1606 #endif
1607   setup_spans_prologue(no)
1608
1609   gvdup_n_u32(x_starts, x_a);
1610   gvcreate_u32(x_ends, x_b, x_c);
1611
1612   setup_spans_down_flat()
1613 }
1614
1615 void setup_spans_up_down(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
1616  vertex_struct *v_b, vertex_struct *v_c)
1617 {
1618 #if 0
1619   setup_spans_up_down_(psx_gpu, v_a, v_b, v_c);
1620   return;
1621 #endif
1622   setup_spans_prologue(no);
1623
1624   s32 y_b = v_b->y;
1625   s64 edge_alt;
1626   s32 edge_dx_dy_alt;
1627   u32 edge_shift_alt;
1628
1629   s32 middle_y = y_a;
1630   s32 height_minor_a = y_a - y_b;
1631   s32 height_minor_b = y_c - y_a;
1632   s32 height_major = y_c - y_b;
1633
1634   vec_2x64s edges_xy_b;
1635   vec_1x64s edges_xy_b_left;
1636   vec_2x32s edges_dx_dy_b;
1637   vec_2x32u edge_shifts_b;
1638
1639   vec_2x32s height_increment;
1640
1641   gvcreate_u32(x_starts, x_a, x_c);
1642   gvdup_n_u32(x_ends, x_b);
1643
1644   compute_edge_delta_x3(x_a, height_minor_a, height_major);
1645
1646   gvcreate_s32(height_increment, 0, height_minor_b);
1647
1648   gvmlal_s32(edges_xy, edges_dx_dy, height_increment);
1649
1650   gvcreate_s64(edges_xy_b_left, edge_alt);
1651   gvcombine_s64(edges_xy_b, edges_xy_b_left, gvhi(edges_xy));
1652
1653   edge_shifts_b = edge_shifts;
1654   gvmov_l_u32(edge_shifts_b, edge_shift_alt, 0);
1655
1656   gvneg_s32(edges_dx_dy_b, edges_dx_dy);
1657   gvmov_l_s32(edges_dx_dy_b, edge_dx_dy_alt, 0);
1658
1659   y_a--;
1660
1661   if(y_b < psx_gpu->viewport_start_y)
1662     height_minor_a -= psx_gpu->viewport_start_y - y_b;
1663
1664   clip = y_a - psx_gpu->viewport_end_y;
1665   if(clip > 0)
1666   {
1667     height_minor_a -= clip;
1668     y_a -= clip;
1669     setup_spans_clip(decrement, no);
1670   }
1671
1672   setup_spans_prologue_b();
1673
1674   if (height_minor_a > 512)
1675     height_minor_a = 512;
1676   if (height_minor_a > 0)
1677   {
1678     u64 y_x4_ = ((u64)(y_a - 3) << 48) | ((u64)(u16)(y_a - 2) << 32)
1679               | (u32)((y_a - 1) << 16) | (u16)y_a;
1680     gvcreate_u64(y_x4, y_x4_);
1681     gvaddw_s32(edges_xy, edges_xy, edges_dx_dy);
1682     setup_spans_adjust_edges_alternate_no(lo, hi);
1683     setup_spans_adjust_interpolants_up();
1684
1685     psx_gpu->num_spans = height_minor_a;
1686     while(height_minor_a > 0)
1687     {
1688       setup_spans_set_x4(none, up, no);
1689       height_minor_a -= 4;
1690     }
1691
1692     span_edge_data += height_minor_a;
1693     span_uvrg_offset += height_minor_a;
1694     span_b_offset += height_minor_a;
1695   }
1696
1697   edges_xy = edges_xy_b;
1698   edges_dx_dy = edges_dx_dy_b;
1699   edge_shifts = edge_shifts_b;
1700
1701   gvld1q_u32(uvrg, psx_gpu->uvrg.e);
1702   b = psx_gpu->b;
1703
1704   y_a = middle_y;
1705
1706   if(y_c > psx_gpu->viewport_end_y)
1707     height_minor_b -= y_c - psx_gpu->viewport_end_y - 1;
1708
1709   clip = psx_gpu->viewport_start_y - y_a;
1710   if(clip > 0)
1711   {
1712     height_minor_b -= clip;
1713     y_a += clip;
1714     setup_spans_clip(increment, no);
1715   }
1716
1717   if (height_minor_b > 512)
1718     height_minor_b = 512;
1719   if (height_minor_b > 0)
1720   {
1721     u64 y_x4_ = ((u64)(y_a + 3) << 48) | ((u64)(u16)(y_a + 2) << 32)
1722               | (u32)((y_a + 1) << 16) | (u16)y_a;
1723     gvcreate_u64(y_x4, y_x4_);
1724     setup_spans_adjust_edges_alternate_no(lo, hi);
1725
1726     // FIXME: overflow corner case
1727     if(psx_gpu->num_spans + height_minor_b == MAX_SPANS)
1728       height_minor_b &= ~3;
1729
1730     psx_gpu->num_spans += height_minor_b;
1731     while(height_minor_b > 0)
1732     {
1733       setup_spans_set_x4(none, down, no);
1734       height_minor_b -= 4;
1735     }
1736     if (psx_gpu->hacks_active & (AHACK_TEXTURE_ADJ_U | AHACK_TEXTURE_ADJ_V))
1737     {
1738       vec_2x32u tmp;
1739       gvld1_u64(tmp, &span_uvrg_offset[height_minor_b - 2]);
1740       gvst1_u64(tmp, &span_uvrg_offset[height_minor_b - 1]);
1741     }
1742   }
1743 }
1744
1745
1746 #define dither_table_entry_normal(value)                                       \
1747   (value)                                                                      \
1748
1749 #define setup_blocks_load_msb_mask_indirect()                                  \
1750
1751 #define setup_blocks_load_msb_mask_direct()                                    \
1752   vec_8x16u msb_mask;                                                          \
1753   gvdupq_n_u16(msb_mask, psx_gpu->mask_msb);                                   \
1754
1755 #define setup_blocks_variables_shaded_textured(target)                         \
1756   vec_4x32u u_block;                                                           \
1757   vec_4x32u v_block;                                                           \
1758   vec_4x32u r_block;                                                           \
1759   vec_4x32u g_block;                                                           \
1760   vec_4x32u b_block;                                                           \
1761   vec_4x32u uvrg_dx;                                                           \
1762   vec_4x32u uvrg_dx4;                                                          \
1763   vec_4x32u uvrg_dx8;                                                          \
1764   vec_4x32u uvrg;                                                              \
1765   vec_16x8u texture_mask;                                                      \
1766   vec_8x8u texture_mask_lo, texture_mask_hi;                                   \
1767   u32 b_dx = psx_gpu->b_block_span.e[1];                                       \
1768   u32 b_dx4 = b_dx << 2;                                                       \
1769   u32 b_dx8 = b_dx << 3;                                                       \
1770   u32 b;                                                                       \
1771                                                                                \
1772   gvld1q_u32(uvrg_dx, psx_gpu->uvrg_dx.e);                                     \
1773   gvshlq_n_u32(uvrg_dx4, uvrg_dx, 2);                                          \
1774   gvshlq_n_u32(uvrg_dx8, uvrg_dx, 3);                                          \
1775   gvld2_u8_dup(texture_mask_lo, texture_mask_hi, &psx_gpu->texture_mask_width); \
1776   gvcombine_u16(texture_mask, texture_mask_lo, texture_mask_hi)                \
1777
1778 #define setup_blocks_variables_shaded_untextured(target)                       \
1779   vec_4x32u r_block;                                                           \
1780   vec_4x32u g_block;                                                           \
1781   vec_4x32u b_block;                                                           \
1782   vec_4x32u rgb_dx;                                                            \
1783   vec_2x32u rgb_dx_lo, rgb_dx_hi;                                              \
1784   vec_4x32u rgb_dx4;                                                           \
1785   vec_4x32u rgb_dx8;                                                           \
1786   vec_4x32u rgb;                                                               \
1787   vec_2x32u rgb_lo, rgb_hi;                                                    \
1788                                                                                \
1789   vec_8x8u d64_0x07;                                                           \
1790   vec_8x8u d64_1;                                                              \
1791   vec_8x8u d64_4;                                                              \
1792   vec_8x8u d64_128;                                                            \
1793                                                                                \
1794   gvdup_n_u8(d64_0x07, 0x07);                                                  \
1795   gvdup_n_u8(d64_1, 1);                                                        \
1796   gvdup_n_u8(d64_4, 4);                                                        \
1797   gvdup_n_u8(d64_128, 128u);                                                   \
1798                                                                                \
1799   gvld1_u32(rgb_dx_lo, &psx_gpu->uvrg_dx.e[2]);                                \
1800   gvcreate_u32(rgb_dx_hi, psx_gpu->b_block_span.e[1], 0);                      \
1801   gvcombine_u32(rgb_dx, rgb_dx_lo, rgb_dx_hi);                                 \
1802   gvshlq_n_u32(rgb_dx4, rgb_dx, 2);                                            \
1803   gvshlq_n_u32(rgb_dx8, rgb_dx, 3)                                             \
1804
1805 #define setup_blocks_variables_unshaded_textured(target)                       \
1806   vec_4x32u u_block;                                                           \
1807   vec_4x32u v_block;                                                           \
1808   vec_2x32u uv_dx;                                                             \
1809   vec_2x32u uv_dx4;                                                            \
1810   vec_2x32u uv_dx8;                                                            \
1811   vec_2x32u uv;                                                                \
1812   vec_16x8u texture_mask;                                                      \
1813   vec_8x8u texture_mask_lo, texture_mask_hi;                                   \
1814                                                                                \
1815   gvld1_u32(uv_dx, psx_gpu->uvrg_dx.e);                                        \
1816   gvld1_u32(uv, psx_gpu->uvrg.e);                                              \
1817   gvshl_n_u32(uv_dx4, uv_dx, 2);                                               \
1818   gvshl_n_u32(uv_dx8, uv_dx, 3);                                               \
1819   gvld2_u8_dup(texture_mask_lo, texture_mask_hi, &psx_gpu->texture_mask_width); \
1820   gvcombine_u16(texture_mask, texture_mask_lo, texture_mask_hi)                \
1821
1822 #define setup_blocks_variables_unshaded_untextured_direct()                    \
1823   gvorrq(colors, colors, msb_mask)                                             \
1824
1825 #define setup_blocks_variables_unshaded_untextured_indirect()                  \
1826
1827 #define setup_blocks_variables_unshaded_untextured(target)                     \
1828   u32 color = psx_gpu->triangle_color;                                         \
1829   vec_8x16u colors;                                                            \
1830                                                                                \
1831   u32 color_r = color & 0xFF;                                                  \
1832   u32 color_g = (color >> 8) & 0xFF;                                           \
1833   u32 color_b = (color >> 16) & 0xFF;                                          \
1834                                                                                \
1835   color = (color_r >> 3) | ((color_g >> 3) << 5) |                             \
1836    ((color_b >> 3) << 10);                                                     \
1837   gvdupq_n_u16(colors, color);                                                 \
1838   setup_blocks_variables_unshaded_untextured_##target()                        \
1839
1840 #define setup_blocks_span_initialize_dithered_textured()                       \
1841   vec_8x16u dither_offsets;                                                    \
1842   gvshll_n_s8(dither_offsets, dither_offsets_short, 4)                         \
1843
1844 #define setup_blocks_span_initialize_dithered_untextured()                     \
1845   vec_8x8u dither_offsets;                                                     \
1846   gvadd_u8(dither_offsets, dither_offsets_short, d64_4)                        \
1847
1848 #define setup_blocks_span_initialize_dithered(texturing)                       \
1849   u32 dither_row = psx_gpu->dither_table[y & 0x3];                             \
1850   u32 dither_shift = (span_edge_data->left_x & 0x3) * 8;                       \
1851   vec_8x8s dither_offsets_short;                                               \
1852                                                                                \
1853   dither_row =                                                                 \
1854    (dither_row >> dither_shift) | (dither_row << (32 - dither_shift));         \
1855   gvdup_n_u32(dither_offsets_short, dither_row);                               \
1856   setup_blocks_span_initialize_dithered_##texturing()                          \
1857
1858 #define setup_blocks_span_initialize_undithered(texturing)                     \
1859
1860 #define setup_blocks_span_initialize_shaded_textured()                         \
1861 {                                                                              \
1862   u32 left_x = span_edge_data->left_x;                                         \
1863   vec_4x32u block_span;                                                        \
1864   vec_4x32u v_left_x;                                                          \
1865                                                                                \
1866   gvld1q_u32(uvrg, span_uvrg_offset);                                          \
1867   gvdupq_n_u32(v_left_x, left_x);                                              \
1868   gvmlaq_u32(uvrg, uvrg_dx, v_left_x);                                         \
1869   b = *span_b_offset;                                                          \
1870   b += b_dx * left_x;                                                          \
1871                                                                                \
1872   gvdupq_l_u32(u_block, gvlo(uvrg), 0);                                        \
1873   gvdupq_l_u32(v_block, gvlo(uvrg), 1);                                        \
1874   gvdupq_l_u32(r_block, gvhi(uvrg), 0);                                        \
1875   gvdupq_l_u32(g_block, gvhi(uvrg), 1);                                        \
1876   gvdupq_n_u32(b_block, b);                                                    \
1877                                                                                \
1878   gvld1q_u32(block_span, psx_gpu->u_block_span.e);                             \
1879   gvaddq_u32(u_block, u_block, block_span);                                    \
1880   gvld1q_u32(block_span, psx_gpu->v_block_span.e);                             \
1881   gvaddq_u32(v_block, v_block, block_span);                                    \
1882   gvld1q_u32(block_span, psx_gpu->r_block_span.e);                             \
1883   gvaddq_u32(r_block, r_block, block_span);                                    \
1884   gvld1q_u32(block_span, psx_gpu->g_block_span.e);                             \
1885   gvaddq_u32(g_block, g_block, block_span);                                    \
1886   gvld1q_u32(block_span, psx_gpu->b_block_span.e);                             \
1887   gvaddq_u32(b_block, b_block, block_span);                                    \
1888 }
1889
1890 #define setup_blocks_span_initialize_shaded_untextured()                       \
1891 {                                                                              \
1892   u32 left_x = span_edge_data->left_x;                                         \
1893   u32 *span_uvrg_offset_high = (u32 *)span_uvrg_offset + 2;                    \
1894   vec_4x32u block_span;                                                        \
1895   vec_4x32u v_left_x;                                                          \
1896                                                                                \
1897   gvld1_u32(rgb_lo, span_uvrg_offset_high);                                    \
1898   gvcreate_u32(rgb_hi, *span_b_offset, 0);                                     \
1899   gvcombine_u32(rgb, rgb_lo, rgb_hi);                                          \
1900   gvdupq_n_u32(v_left_x, left_x);                                              \
1901   gvmlaq_u32(rgb, rgb_dx, v_left_x);                                           \
1902                                                                                \
1903   gvdupq_l_u32(r_block, gvlo(rgb), 0);                                         \
1904   gvdupq_l_u32(g_block, gvlo(rgb), 1);                                         \
1905   gvdupq_l_u32(b_block, gvhi(rgb), 0);                                         \
1906                                                                                \
1907   gvld1q_u32(block_span, psx_gpu->r_block_span.e);                             \
1908   gvaddq_u32(r_block, r_block, block_span);                                    \
1909   gvld1q_u32(block_span, psx_gpu->g_block_span.e);                             \
1910   gvaddq_u32(g_block, g_block, block_span);                                    \
1911   gvld1q_u32(block_span, psx_gpu->b_block_span.e);                             \
1912   gvaddq_u32(b_block, b_block, block_span);                                    \
1913 }                                                                              \
1914
1915 #define setup_blocks_span_initialize_unshaded_textured()                       \
1916 {                                                                              \
1917   u32 left_x = span_edge_data->left_x;                                         \
1918   vec_4x32u block_span;                                                        \
1919   vec_2x32u v_left_x;                                                          \
1920                                                                                \
1921   gvld1_u32(uv, span_uvrg_offset);                                             \
1922   gvdup_n_u32(v_left_x, left_x);                                               \
1923   gvmla_u32(uv, uv_dx, v_left_x);                                              \
1924                                                                                \
1925   gvdupq_l_u32(u_block, uv, 0);                                                \
1926   gvdupq_l_u32(v_block, uv, 1);                                                \
1927                                                                                \
1928   gvld1q_u32(block_span, psx_gpu->u_block_span.e);                             \
1929   gvaddq_u32(u_block, u_block, block_span);                                    \
1930   gvld1q_u32(block_span, psx_gpu->v_block_span.e);                             \
1931   gvaddq_u32(v_block, v_block, block_span);                                    \
1932 }                                                                              \
1933
1934 #define setup_blocks_span_initialize_unshaded_untextured()                     \
1935
1936 #define setup_blocks_texture_swizzled()                                        \
1937 {                                                                              \
1938   vec_8x8u u_saved = u;                                                        \
1939   gvsli_n_u8(u, v, 4);                                                         \
1940   gvsri_n_u8(v, u_saved, 4);                                                   \
1941 }                                                                              \
1942
1943 #define setup_blocks_texture_unswizzled()                                      \
1944
1945 #define setup_blocks_store_shaded_textured(swizzling, dithering, target,       \
1946  edge_type)                                                                    \
1947 {                                                                              \
1948   vec_8x16u u_whole;                                                           \
1949   vec_8x16u v_whole;                                                           \
1950   vec_8x16u r_whole;                                                           \
1951   vec_8x16u g_whole;                                                           \
1952   vec_8x16u b_whole;                                                           \
1953   vec_4x16u u_whole_lo, u_whole_hi;                                            \
1954   vec_4x16u v_whole_lo, v_whole_hi;                                            \
1955   vec_4x16u r_whole_lo, r_whole_hi;                                            \
1956   vec_4x16u g_whole_lo, g_whole_hi;                                            \
1957   vec_4x16u b_whole_lo, b_whole_hi;                                            \
1958                                                                                \
1959   vec_8x8u u;                                                                  \
1960   vec_8x8u v;                                                                  \
1961   vec_8x8u r;                                                                  \
1962   vec_8x8u g;                                                                  \
1963   vec_8x8u b;                                                                  \
1964                                                                                \
1965   vec_4x32u dx4;                                                               \
1966   vec_4x32u dx8;                                                               \
1967                                                                                \
1968   gvshrn_n_u32(u_whole_lo, u_block, 16);                                       \
1969   gvshrn_n_u32(v_whole_lo, v_block, 16);                                       \
1970   gvshrn_n_u32(r_whole_lo, r_block, 16);                                       \
1971   gvshrn_n_u32(g_whole_lo, g_block, 16);                                       \
1972   gvshrn_n_u32(b_whole_lo, b_block, 16);                                       \
1973                                                                                \
1974   gvdupq_l_u32(dx4, gvlo(uvrg_dx4), 0);                                        \
1975   gvaddhn_u32(u_whole_hi, u_block, dx4);                                       \
1976   gvdupq_l_u32(dx4, gvlo(uvrg_dx4), 1);                                        \
1977   gvaddhn_u32(v_whole_hi, v_block, dx4);                                       \
1978   gvdupq_l_u32(dx4, gvhi(uvrg_dx4), 0);                                        \
1979   gvaddhn_u32(r_whole_hi, r_block, dx4);                                       \
1980   gvdupq_l_u32(dx4, gvhi(uvrg_dx4), 1);                                        \
1981   gvaddhn_u32(g_whole_hi, g_block, dx4);                                       \
1982   gvdupq_n_u32(dx4, b_dx4);                                                    \
1983   gvaddhn_u32(b_whole_hi, b_block, dx4);                                       \
1984                                                                                \
1985   gvcombine_u16(u_whole, u_whole_lo, u_whole_hi);                              \
1986   gvcombine_u16(v_whole, v_whole_lo, v_whole_hi);                              \
1987   gvcombine_u16(r_whole, r_whole_lo, r_whole_hi);                              \
1988   gvcombine_u16(g_whole, g_whole_lo, g_whole_hi);                              \
1989   gvcombine_u16(b_whole, b_whole_lo, b_whole_hi);                              \
1990   gvmovn_u16(u, u_whole);                                                      \
1991   gvmovn_u16(v, v_whole);                                                      \
1992   gvmovn_u16(r, r_whole);                                                      \
1993   gvmovn_u16(g, g_whole);                                                      \
1994   gvmovn_u16(b, b_whole);                                                      \
1995                                                                                \
1996   gvdupq_l_u32(dx8, gvlo(uvrg_dx8), 0);                                        \
1997   gvaddq_u32(u_block, u_block, dx8);                                           \
1998   gvdupq_l_u32(dx8, gvlo(uvrg_dx8), 1);                                        \
1999   gvaddq_u32(v_block, v_block, dx8);                                           \
2000   gvdupq_l_u32(dx8, gvhi(uvrg_dx8), 0);                                        \
2001   gvaddq_u32(r_block, r_block, dx8);                                           \
2002   gvdupq_l_u32(dx8, gvhi(uvrg_dx8), 1);                                        \
2003   gvaddq_u32(g_block, g_block, dx8);                                           \
2004   gvdupq_n_u32(dx8, b_dx8);                                                    \
2005   gvaddq_u32(b_block, b_block, dx8);                                           \
2006                                                                                \
2007   gvand(u, u, gvlo(texture_mask));                                             \
2008   gvand(v, v, gvhi(texture_mask));                                             \
2009   setup_blocks_texture_##swizzling();                                          \
2010                                                                                \
2011   gvst2_u8(u, v, (u8 *)block->uv.e);                                           \
2012   gvst1_u8(r, block->r.e);                                                     \
2013   gvst1_u8(g, block->g.e);                                                     \
2014   gvst1_u8(b, block->b.e);                                                     \
2015   gvst1q_u16(dither_offsets, (u16 *)block->dither_offsets.e);                  \
2016   block->fb_ptr = fb_ptr;                                                      \
2017 }                                                                              \
2018
2019 #define setup_blocks_store_unshaded_textured(swizzling, dithering, target,     \
2020  edge_type)                                                                    \
2021 {                                                                              \
2022   vec_8x16u u_whole;                                                           \
2023   vec_8x16u v_whole;                                                           \
2024   vec_4x16u u_whole_lo, u_whole_hi;                                            \
2025   vec_4x16u v_whole_lo, v_whole_hi;                                            \
2026                                                                                \
2027   vec_8x8u u;                                                                  \
2028   vec_8x8u v;                                                                  \
2029                                                                                \
2030   vec_4x32u dx4;                                                               \
2031   vec_4x32u dx8;                                                               \
2032                                                                                \
2033   gvshrn_n_u32(u_whole_lo, u_block, 16);                                       \
2034   gvshrn_n_u32(v_whole_lo, v_block, 16);                                       \
2035                                                                                \
2036   gvdupq_l_u32(dx4, uv_dx4, 0);                                                \
2037   gvaddhn_u32(u_whole_hi, u_block, dx4);                                       \
2038   gvdupq_l_u32(dx4, uv_dx4, 1);                                                \
2039   gvaddhn_u32(v_whole_hi, v_block, dx4);                                       \
2040                                                                                \
2041   gvcombine_u16(u_whole, u_whole_lo, u_whole_hi);                              \
2042   gvcombine_u16(v_whole, v_whole_lo, v_whole_hi);                              \
2043   gvmovn_u16(u, u_whole);                                                      \
2044   gvmovn_u16(v, v_whole);                                                      \
2045                                                                                \
2046   gvdupq_l_u32(dx8, uv_dx8, 0);                                                \
2047   gvaddq_u32(u_block, u_block, dx8);                                           \
2048   gvdupq_l_u32(dx8, uv_dx8, 1);                                                \
2049   gvaddq_u32(v_block, v_block, dx8);                                           \
2050                                                                                \
2051   gvand(u, u, gvlo(texture_mask));                                             \
2052   gvand(v, v, gvhi(texture_mask));                                             \
2053   setup_blocks_texture_##swizzling();                                          \
2054                                                                                \
2055   gvst2_u8(u, v, (u8 *)block->uv.e);                                           \
2056   gvst1q_u16(dither_offsets, (u16 *)block->dither_offsets.e);                  \
2057   block->fb_ptr = fb_ptr;                                                      \
2058 }                                                                              \
2059
2060 #define setup_blocks_store_shaded_untextured_dithered()                        \
2061   gvqadd_u8(r, r, dither_offsets);                                             \
2062   gvqadd_u8(g, g, dither_offsets);                                             \
2063   gvqadd_u8(b, b, dither_offsets);                                             \
2064                                                                                \
2065   gvqsub_u8(r, r, d64_4);                                                      \
2066   gvqsub_u8(g, g, d64_4);                                                      \
2067   gvqsub_u8(b, b, d64_4)                                                       \
2068
2069 #define setup_blocks_store_shaded_untextured_undithered()                      \
2070
2071 #define setup_blocks_store_untextured_pixels_indirect_full(_pixels)            \
2072   gvst1q_u16(_pixels, block->pixels.e);                                        \
2073   block->fb_ptr = fb_ptr                                                       \
2074
2075 #define setup_blocks_store_untextured_pixels_indirect_edge(_pixels)            \
2076   gvst1q_u16(_pixels, block->pixels.e);                                        \
2077   block->fb_ptr = fb_ptr                                                       \
2078
2079 #define setup_blocks_store_shaded_untextured_seed_pixels_indirect()            \
2080   gvmull_u8(pixels, r, d64_1)                                                  \
2081
2082 #define setup_blocks_store_untextured_pixels_direct_full(_pixels)              \
2083   gvst1q_u16(_pixels, fb_ptr)                                                  \
2084
2085 #define setup_blocks_store_untextured_pixels_direct_edge(_pixels)              \
2086 {                                                                              \
2087   vec_8x16u fb_pixels;                                                         \
2088   vec_8x16u draw_mask;                                                         \
2089   vec_8x16u test_mask;                                                         \
2090                                                                                \
2091   gvld1q_u16(test_mask, psx_gpu->test_mask.e);                                 \
2092   gvld1q_u16(fb_pixels, fb_ptr);                                               \
2093   gvdupq_n_u16(draw_mask, span_edge_data->right_mask);                         \
2094   gvtstq_u16(draw_mask, draw_mask, test_mask);                                 \
2095   gvbifq(fb_pixels, _pixels, draw_mask);                                       \
2096   gvst1q_u16(fb_pixels, fb_ptr);                                               \
2097 }                                                                              \
2098
2099 #define setup_blocks_store_shaded_untextured_seed_pixels_direct()              \
2100   pixels = msb_mask;                                                           \
2101   gvmlal_u8(pixels, r, d64_1)                                                  \
2102
2103 #define setup_blocks_store_shaded_untextured(swizzling, dithering, target,     \
2104  edge_type)                                                                    \
2105 {                                                                              \
2106   vec_8x16u r_whole;                                                           \
2107   vec_8x16u g_whole;                                                           \
2108   vec_8x16u b_whole;                                                           \
2109   vec_4x16u r_whole_lo, r_whole_hi;                                            \
2110   vec_4x16u g_whole_lo, g_whole_hi;                                            \
2111   vec_4x16u b_whole_lo, b_whole_hi;                                            \
2112                                                                                \
2113   vec_8x8u r;                                                                  \
2114   vec_8x8u g;                                                                  \
2115   vec_8x8u b;                                                                  \
2116                                                                                \
2117   vec_4x32u dx4;                                                               \
2118   vec_4x32u dx8;                                                               \
2119                                                                                \
2120   vec_8x16u pixels;                                                            \
2121                                                                                \
2122   gvshrn_n_u32(r_whole_lo, r_block, 16);                                       \
2123   gvshrn_n_u32(g_whole_lo, g_block, 16);                                       \
2124   gvshrn_n_u32(b_whole_lo, b_block, 16);                                       \
2125                                                                                \
2126   gvdupq_l_u32(dx4, gvlo(rgb_dx4), 0);                                         \
2127   gvaddhn_u32(r_whole_hi, r_block, dx4);                                       \
2128   gvdupq_l_u32(dx4, gvlo(rgb_dx4), 1);                                         \
2129   gvaddhn_u32(g_whole_hi, g_block, dx4);                                       \
2130   gvdupq_l_u32(dx4, gvhi(rgb_dx4), 0);                                         \
2131   gvaddhn_u32(b_whole_hi, b_block, dx4);                                       \
2132                                                                                \
2133   gvcombine_u16(r_whole, r_whole_lo, r_whole_hi);                              \
2134   gvcombine_u16(g_whole, g_whole_lo, g_whole_hi);                              \
2135   gvcombine_u16(b_whole, b_whole_lo, b_whole_hi);                              \
2136   gvmovn_u16(r, r_whole);                                                      \
2137   gvmovn_u16(g, g_whole);                                                      \
2138   gvmovn_u16(b, b_whole);                                                      \
2139                                                                                \
2140   gvdupq_l_u32(dx8, gvlo(rgb_dx8), 0);                                         \
2141   gvaddq_u32(r_block, r_block, dx8);                                           \
2142   gvdupq_l_u32(dx8, gvlo(rgb_dx8), 1);                                         \
2143   gvaddq_u32(g_block, g_block, dx8);                                           \
2144   gvdupq_l_u32(dx8, gvhi(rgb_dx8), 0);                                         \
2145   gvaddq_u32(b_block, b_block, dx8);                                           \
2146                                                                                \
2147   setup_blocks_store_shaded_untextured_##dithering();                          \
2148                                                                                \
2149   gvshr_n_u8(r, r, 3);                                                         \
2150   gvbic(g, g, d64_0x07);                                                       \
2151   gvbic(b, b, d64_0x07);                                                       \
2152                                                                                \
2153   setup_blocks_store_shaded_untextured_seed_pixels_##target();                 \
2154   gvmlal_u8(pixels, g, d64_4);                                                 \
2155   gvmlal_u8(pixels, b, d64_128);                                               \
2156                                                                                \
2157   setup_blocks_store_untextured_pixels_##target##_##edge_type(pixels);         \
2158 }                                                                              \
2159
2160 #define setup_blocks_store_unshaded_untextured(swizzling, dithering, target,   \
2161  edge_type)                                                                    \
2162   setup_blocks_store_untextured_pixels_##target##_##edge_type(colors)          \
2163
2164 #define setup_blocks_store_draw_mask_textured_indirect(_block, bits)           \
2165   (_block)->draw_mask_bits = bits                                              \
2166
2167 #define setup_blocks_store_draw_mask_untextured_indirect(_block, bits)         \
2168 {                                                                              \
2169   vec_8x16u bits_mask;                                                         \
2170   vec_8x16u test_mask;                                                         \
2171                                                                                \
2172   gvld1q_u16(test_mask, psx_gpu->test_mask.e);                                 \
2173   gvdupq_n_u16(bits_mask, bits);                                               \
2174   gvtstq_u16(bits_mask, bits_mask, test_mask);                                 \
2175   gvst1q_u16(bits_mask, (_block)->draw_mask.e);                                \
2176 }                                                                              \
2177
2178 #define setup_blocks_store_draw_mask_untextured_direct(_block, bits)           \
2179
2180 #define setup_blocks_uv_adj_hack_untextured(_block, edge_data, uvrg_offset)    \
2181
2182 #define setup_blocks_uv_adj_hack_textured(_block, edge_data, uvrg_offset)      \
2183 {                                                                              \
2184   u32 m_ = AHACK_TEXTURE_ADJ_U | AHACK_TEXTURE_ADJ_V;                          \
2185   if (unlikely(psx_gpu->hacks_active & m_))                                    \
2186     setup_blocks_uv_adj_hack(psx_gpu, _block, edge_data, (void *)uvrg_offset); \
2187 }                                                                              \
2188
2189 #define setup_blocks_add_blocks_indirect()                                     \
2190   num_blocks += span_num_blocks;                                               \
2191                                                                                \
2192   if(num_blocks > MAX_BLOCKS)                                                  \
2193   {                                                                            \
2194     psx_gpu->num_blocks = num_blocks - span_num_blocks;                        \
2195     flush_render_block_buffer(psx_gpu);                                        \
2196     num_blocks = span_num_blocks;                                              \
2197     block = psx_gpu->blocks;                                                   \
2198   }                                                                            \
2199
2200 #define setup_blocks_add_blocks_direct()                                       \
2201
2202 #define setup_blocks_do(shading, texturing, dithering, sw, target)             \
2203   setup_blocks_load_msb_mask_##target();                                       \
2204   setup_blocks_variables_##shading##_##texturing(target);                      \
2205                                                                                \
2206   edge_data_struct *span_edge_data = psx_gpu->span_edge_data;                  \
2207   vec_4x32u *span_uvrg_offset = (vec_4x32u *)psx_gpu->span_uvrg_offset;        \
2208   u32 *span_b_offset = psx_gpu->span_b_offset;                                 \
2209                                                                                \
2210   block_struct *block = psx_gpu->blocks + psx_gpu->num_blocks;                 \
2211                                                                                \
2212   u32 num_spans = psx_gpu->num_spans;                                          \
2213                                                                                \
2214   u16 * __restrict__ fb_ptr;                                                   \
2215   u32 y;                                                                       \
2216                                                                                \
2217   u32 num_blocks = psx_gpu->num_blocks;                                        \
2218   u32 span_num_blocks;                                                         \
2219                                                                                \
2220   while(num_spans)                                                             \
2221   {                                                                            \
2222     span_num_blocks = span_edge_data->num_blocks;                              \
2223     if(span_num_blocks)                                                        \
2224     {                                                                          \
2225       y = span_edge_data->y;                                                   \
2226       fb_ptr = psx_gpu->vram_out_ptr + span_edge_data->left_x + (y * 1024);    \
2227                                                                                \
2228       setup_blocks_span_initialize_##shading##_##texturing();                  \
2229       setup_blocks_span_initialize_##dithering(texturing);                     \
2230                                                                                \
2231       setup_blocks_add_blocks_##target();                                      \
2232                                                                                \
2233       span_num_blocks--;                                                       \
2234       while(span_num_blocks)                                                   \
2235       {                                                                        \
2236         setup_blocks_store_##shading##_##texturing(sw, dithering, target,      \
2237          full);                                                                \
2238         setup_blocks_store_draw_mask_##texturing##_##target(block, 0x00);      \
2239                                                                                \
2240         fb_ptr += 8;                                                           \
2241         block++;                                                               \
2242         span_num_blocks--;                                                     \
2243       }                                                                        \
2244                                                                                \
2245       setup_blocks_store_##shading##_##texturing(sw, dithering, target, edge); \
2246       setup_blocks_store_draw_mask_##texturing##_##target(block,               \
2247        span_edge_data->right_mask);                                            \
2248       setup_blocks_uv_adj_hack_##texturing(block, span_edge_data,              \
2249           span_uvrg_offset);                                                   \
2250                                                                                \
2251       block++;                                                                 \
2252     }                                                                          \
2253                                                                                \
2254     num_spans--;                                                               \
2255     span_edge_data++;                                                          \
2256     span_uvrg_offset++;                                                        \
2257     span_b_offset++;                                                           \
2258   }                                                                            \
2259                                                                                \
2260   psx_gpu->num_blocks = num_blocks                                             \
2261
2262 void setup_blocks_shaded_textured_dithered_swizzled_indirect(psx_gpu_struct
2263  *psx_gpu)
2264 {
2265 #if 0
2266   setup_blocks_shaded_textured_dithered_swizzled_indirect_(psx_gpu);
2267   return;
2268 #endif
2269   setup_blocks_do(shaded, textured, dithered, swizzled, indirect);
2270 }
2271
2272 void setup_blocks_shaded_textured_dithered_unswizzled_indirect(psx_gpu_struct
2273  *psx_gpu)
2274 {
2275 #if 0
2276   setup_blocks_shaded_textured_dithered_unswizzled_indirect_(psx_gpu);
2277   return;
2278 #endif
2279   setup_blocks_do(shaded, textured, dithered, unswizzled, indirect);
2280 }
2281
2282 void setup_blocks_unshaded_textured_dithered_swizzled_indirect(psx_gpu_struct
2283  *psx_gpu)
2284 {
2285 #if 0
2286   setup_blocks_unshaded_textured_dithered_swizzled_indirect_(psx_gpu);
2287   return;
2288 #endif
2289   setup_blocks_do(unshaded, textured, dithered, swizzled, indirect);
2290 }
2291
2292 void setup_blocks_unshaded_textured_dithered_unswizzled_indirect(psx_gpu_struct
2293  *psx_gpu)
2294 {
2295 #if 0
2296   setup_blocks_unshaded_textured_dithered_unswizzled_indirect_(psx_gpu);
2297   return;
2298 #endif
2299   setup_blocks_do(unshaded, textured, dithered, unswizzled, indirect);
2300 }
2301
2302 void setup_blocks_unshaded_untextured_undithered_unswizzled_indirect(
2303  psx_gpu_struct *psx_gpu)
2304 {
2305 #if 0
2306   setup_blocks_unshaded_untextured_undithered_unswizzled_indirect_(psx_gpu);
2307   return;
2308 #endif
2309   setup_blocks_do(unshaded, untextured, undithered, unswizzled, indirect);
2310 }
2311
2312 void setup_blocks_unshaded_untextured_undithered_unswizzled_direct(
2313  psx_gpu_struct *psx_gpu)
2314 {
2315 #if 0
2316   setup_blocks_unshaded_untextured_undithered_unswizzled_direct_(psx_gpu);
2317   return;
2318 #endif
2319   setup_blocks_do(unshaded, untextured, undithered, unswizzled, direct);
2320 }
2321
2322 void setup_blocks_shaded_untextured_undithered_unswizzled_indirect(psx_gpu_struct
2323  *psx_gpu)
2324 {
2325 #if 0
2326   setup_blocks_shaded_untextured_undithered_unswizzled_indirect_(psx_gpu);
2327   return;
2328 #endif
2329   setup_blocks_do(shaded, untextured, undithered, unswizzled, indirect);
2330 }
2331
2332 void setup_blocks_shaded_untextured_dithered_unswizzled_indirect(psx_gpu_struct
2333  *psx_gpu)
2334 {
2335 #if 0
2336   setup_blocks_shaded_untextured_dithered_unswizzled_indirect_(psx_gpu);
2337   return;
2338 #endif
2339   setup_blocks_do(shaded, untextured, dithered, unswizzled, indirect);
2340 }
2341
2342 void setup_blocks_shaded_untextured_undithered_unswizzled_direct(
2343  psx_gpu_struct *psx_gpu)
2344 {
2345 #if 0
2346   setup_blocks_shaded_untextured_undithered_unswizzled_direct_(psx_gpu);
2347   return;
2348 #endif
2349   setup_blocks_do(shaded, untextured, undithered, unswizzled, direct);
2350 }
2351
2352 void setup_blocks_shaded_untextured_dithered_unswizzled_direct(psx_gpu_struct
2353  *psx_gpu)
2354 {
2355 #if 0
2356   setup_blocks_shaded_untextured_dithered_unswizzled_direct_(psx_gpu);
2357   return;
2358 #endif
2359   setup_blocks_do(shaded, untextured, dithered, unswizzled, direct);
2360 }
2361
2362 static void update_texture_4bpp_cache(psx_gpu_struct *psx_gpu)
2363 {
2364   u32 current_texture_page = psx_gpu->current_texture_page;
2365   u8 *texture_page_ptr = psx_gpu->texture_page_base;
2366   const u16 *vram_ptr = psx_gpu->vram_ptr;
2367   u32 tile_x, tile_y;
2368   u32 sub_y;
2369   vec_8x16u c_0x00f0;
2370
2371   vram_ptr += (current_texture_page >> 4) * 256 * 1024;
2372   vram_ptr += (current_texture_page & 0xF) * 64;
2373
2374   gvdupq_n_u16(c_0x00f0, 0x00f0);
2375
2376   psx_gpu->dirty_textures_4bpp_mask &= ~(psx_gpu->current_texture_mask);
2377
2378   for (tile_y = 16; tile_y; tile_y--)
2379   {
2380     for (tile_x = 16; tile_x; tile_x--)
2381     {
2382       for (sub_y = 8; sub_y; sub_y--)
2383       {
2384         vec_8x8u texel_block_a, texel_block_b;
2385         vec_8x16u texel_block_expanded_a, texel_block_expanded_b;
2386         vec_8x16u texel_block_expanded_c, texel_block_expanded_d;
2387         vec_8x16u texel_block_expanded_ab, texel_block_expanded_cd;
2388
2389         gvld1_u8(texel_block_a, (u8 *)vram_ptr); vram_ptr += 1024;
2390         gvld1_u8(texel_block_b, (u8 *)vram_ptr); vram_ptr += 1024;
2391
2392         gvmovl_u8(texel_block_expanded_a, texel_block_a);
2393         gvshll_n_u8(texel_block_expanded_b, texel_block_a, 4);
2394         gvmovl_u8(texel_block_expanded_c, texel_block_b);
2395         gvshll_n_u8(texel_block_expanded_d, texel_block_b, 4);
2396
2397         gvbicq(texel_block_expanded_a, texel_block_expanded_a, c_0x00f0);
2398         gvbicq(texel_block_expanded_b, texel_block_expanded_b, c_0x00f0);
2399         gvbicq(texel_block_expanded_c, texel_block_expanded_c, c_0x00f0);
2400         gvbicq(texel_block_expanded_d, texel_block_expanded_d, c_0x00f0);
2401
2402         gvorrq(texel_block_expanded_ab, texel_block_expanded_a, texel_block_expanded_b);
2403         gvorrq(texel_block_expanded_cd, texel_block_expanded_c, texel_block_expanded_d);
2404
2405         gvst1q_2_pi_u32(texel_block_expanded_ab, texel_block_expanded_cd, texture_page_ptr);
2406       }
2407
2408       vram_ptr -= (1024 * 16) - 4;
2409     }
2410
2411     vram_ptr += (16 * 1024) - (4 * 16);
2412   }
2413 }
2414
2415 void update_texture_8bpp_cache_slice(psx_gpu_struct *psx_gpu,
2416  u32 texture_page)
2417 {
2418 #if 0
2419   update_texture_8bpp_cache_slice_(psx_gpu, texture_page);
2420   return;
2421 #endif
2422   u16 *texture_page_ptr = psx_gpu->texture_page_base;
2423   u16 *vram_ptr = psx_gpu->vram_ptr;
2424
2425   u32 tile_x, tile_y;
2426   u32 sub_y;
2427
2428   vram_ptr += (texture_page >> 4) * 256 * 1024;
2429   vram_ptr += (texture_page & 0xF) * 64;
2430
2431   if((texture_page ^ psx_gpu->current_texture_page) & 0x1)
2432     texture_page_ptr += (8 * 16) * 8;
2433
2434   for (tile_y = 16; tile_y; tile_y--)
2435   {
2436     for (tile_x = 8; tile_x; tile_x--)
2437     {
2438       for (sub_y = 4; sub_y; sub_y--)
2439       {
2440         vec_4x32u texels_a, texels_b, texels_c, texels_d = {};
2441         gvld1q_u32(texels_a, vram_ptr); vram_ptr += 1024;
2442         gvld1q_u32(texels_b, vram_ptr); vram_ptr += 1024;
2443         gvld1q_u32(texels_c, vram_ptr); vram_ptr += 1024;
2444         gvld1q_u32(texels_d, vram_ptr); vram_ptr += 1024;
2445
2446         gvst1q_2_pi_u32(texels_a, texels_b, texture_page_ptr);
2447         gvst1q_2_pi_u32(texels_c, texels_d, texture_page_ptr);
2448       }
2449
2450       vram_ptr -= (1024 * 16) - 8;
2451     }
2452
2453     vram_ptr -= (8 * 8);
2454     vram_ptr += (16 * 1024);
2455
2456     texture_page_ptr += (8 * 16) * 8;
2457   }
2458 }
2459
2460 void texture_blocks_untextured(psx_gpu_struct *psx_gpu)
2461 {
2462 }
2463
2464 void texture_blocks_4bpp(psx_gpu_struct *psx_gpu)
2465 {
2466 #if 0
2467   texture_blocks_4bpp_(psx_gpu);
2468   return;
2469 #endif
2470   block_struct *block = psx_gpu->blocks;
2471   u32 num_blocks = psx_gpu->num_blocks;
2472
2473   vec_8x8u texels_low;
2474   vec_8x8u texels_high;
2475
2476   vec_16x8u clut_low;
2477   vec_16x8u clut_high;
2478
2479   const u8 *texture_ptr_8bpp = psx_gpu->texture_page_ptr;
2480
2481   gvld2q_u8(clut_low, clut_high, (u8 *)psx_gpu->clut_ptr);
2482
2483   if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_4bpp_mask)
2484     update_texture_4bpp_cache(psx_gpu);
2485
2486   while(num_blocks)
2487   {
2488     vec_8x8u texels =
2489     {
2490       .u8 =
2491       {
2492         texture_ptr_8bpp[block->uv.e[0]],
2493         texture_ptr_8bpp[block->uv.e[1]],
2494         texture_ptr_8bpp[block->uv.e[2]],
2495         texture_ptr_8bpp[block->uv.e[3]],
2496         texture_ptr_8bpp[block->uv.e[4]],
2497         texture_ptr_8bpp[block->uv.e[5]],
2498         texture_ptr_8bpp[block->uv.e[6]],
2499         texture_ptr_8bpp[block->uv.e[7]]
2500       }
2501     };
2502
2503     gvtbl2_u8(texels_low, clut_low, texels);
2504     gvtbl2_u8(texels_high, clut_high, texels);
2505
2506     gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e);
2507
2508     num_blocks--;
2509     block++;
2510   }
2511 }
2512
2513 void texture_blocks_8bpp(psx_gpu_struct *psx_gpu)
2514 {
2515 #if 0
2516   texture_blocks_8bpp_(psx_gpu);
2517   return;
2518 #endif
2519   u32 num_blocks = psx_gpu->num_blocks;
2520
2521   if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_8bpp_mask)
2522     update_texture_8bpp_cache(psx_gpu);
2523
2524   const u8 * __restrict__ texture_ptr_8bpp = psx_gpu->texture_page_ptr;
2525   const u16 * __restrict__ clut_ptr = psx_gpu->clut_ptr;
2526   block_struct * __restrict__ block = psx_gpu->blocks;
2527
2528   while(num_blocks)
2529   {
2530     u16 offset;
2531     #define load_one(i_) \
2532       offset = block->uv.e[i_]; u16 texel##i_ = texture_ptr_8bpp[offset]
2533     #define store_one(i_) \
2534       block->texels.e[i_] = clut_ptr[texel##i_]
2535     load_one(0); load_one(1); load_one(2); load_one(3);
2536     load_one(4); load_one(5); load_one(6); load_one(7);
2537     store_one(0); store_one(1); store_one(2); store_one(3);
2538     store_one(4); store_one(5); store_one(6); store_one(7);
2539     #undef load_one
2540     #undef store_one
2541
2542     num_blocks--;
2543     block++;
2544   }
2545 }
2546
2547 void texture_blocks_16bpp(psx_gpu_struct *psx_gpu)
2548 {
2549 #if 0
2550   texture_blocks_16bpp_(psx_gpu);
2551   return;
2552 #endif
2553   u32 num_blocks = psx_gpu->num_blocks;
2554   const u16 * __restrict__ texture_ptr_16bpp = psx_gpu->texture_page_ptr;
2555   block_struct * __restrict__ block = psx_gpu->blocks;
2556
2557   while(num_blocks)
2558   {
2559     u32 offset;
2560     #define load_one(i_) \
2561       offset = block->uv.e[i_]; \
2562       offset += ((offset & 0xFF00) * 3); \
2563       u16 texel##i_ = texture_ptr_16bpp[offset]
2564     #define store_one(i_) \
2565       block->texels.e[i_] = texel##i_
2566     load_one(0); load_one(1); load_one(2); load_one(3);
2567     load_one(4); load_one(5); load_one(6); load_one(7);
2568     store_one(0); store_one(1); store_one(2); store_one(3);
2569     store_one(4); store_one(5); store_one(6); store_one(7);
2570     #undef load_one
2571     #undef store_one
2572
2573     num_blocks--;
2574     block++;
2575   }
2576 }
2577
2578 #define shade_blocks_load_msb_mask_indirect()                                  \
2579
2580 #define shade_blocks_load_msb_mask_direct()                                    \
2581   vec_8x16u msb_mask;                                                          \
2582   gvdupq_n_u16(msb_mask, psx_gpu->mask_msb);                                   \
2583
2584 #define shade_blocks_store_indirect(_draw_mask, _pixels)                       \
2585   gvst1q_u16(_draw_mask, block->draw_mask.e);                                  \
2586   gvst1q_u16(_pixels, block->pixels.e);                                        \
2587
2588 #define shade_blocks_store_direct(_draw_mask, _pixels)                         \
2589 {                                                                              \
2590   u16 * __restrict__ fb_ptr = block->fb_ptr;                                   \
2591   vec_8x16u fb_pixels;                                                         \
2592   gvld1q_u16(fb_pixels, fb_ptr);                                               \
2593   gvorrq(_pixels, _pixels, msb_mask);                                          \
2594   gvbifq(fb_pixels, _pixels, _draw_mask);                                      \
2595   gvst1q_u16(fb_pixels, fb_ptr);                                               \
2596 }                                                                              \
2597
2598 #define shade_blocks_textured_false_modulated_check_dithered(target)           \
2599
2600 #define shade_blocks_textured_false_modulated_check_undithered(target)         \
2601   if(psx_gpu->triangle_color == 0x808080)                                      \
2602   {                                                                            \
2603     shade_blocks_textured_unmodulated_##target(psx_gpu);                       \
2604     return;                                                                    \
2605   }                                                                            \
2606
2607 #define shade_blocks_textured_modulated_shaded_primitive_load(dithering,       \
2608  target)                                                                       \
2609
2610 #define shade_blocks_textured_modulated_unshaded_primitive_load(dithering,     \
2611  target)                                                                       \
2612 {                                                                              \
2613   u32 color = psx_gpu->triangle_color;                                         \
2614   gvdup_n_u8(colors_r, color);                                                 \
2615   gvdup_n_u8(colors_g, color >> 8);                                            \
2616   gvdup_n_u8(colors_b, color >> 16);                                           \
2617   shade_blocks_textured_false_modulated_check_##dithering(target);             \
2618 }                                                                              \
2619
2620 #define shade_blocks_textured_modulated_shaded_block_load()                    \
2621   gvld1_u8(colors_r, block->r.e);                                              \
2622   gvld1_u8(colors_g, block->g.e);                                              \
2623   gvld1_u8(colors_b, block->b.e)                                               \
2624
2625 #define shade_blocks_textured_modulated_unshaded_block_load()                  \
2626
2627 #define shade_blocks_textured_modulate_dithered(component)                     \
2628   gvld1q_u16(pixels_##component, block->dither_offsets.e);                     \
2629   gvmlal_u8(pixels_##component, texels_##component, colors_##component)        \
2630
2631 #define shade_blocks_textured_modulate_undithered(component)                   \
2632   gvmull_u8(pixels_##component, texels_##component, colors_##component)        \
2633
2634 #define shade_blocks_textured_modulated_do(shading, dithering, target)         \
2635   block_struct * __restrict__ block = psx_gpu->blocks;                         \
2636   u32 num_blocks = psx_gpu->num_blocks;                                        \
2637   vec_8x16u texels;                                                            \
2638                                                                                \
2639   vec_8x8u texels_r;                                                           \
2640   vec_8x8u texels_g;                                                           \
2641   vec_8x8u texels_b;                                                           \
2642                                                                                \
2643   vec_8x8u colors_r;                                                           \
2644   vec_8x8u colors_g;                                                           \
2645   vec_8x8u colors_b;                                                           \
2646                                                                                \
2647   vec_8x8u pixels_r_low;                                                       \
2648   vec_8x8u pixels_g_low;                                                       \
2649   vec_8x8u pixels_b_low;                                                       \
2650   vec_8x16u pixels;                                                            \
2651                                                                                \
2652   vec_8x16u pixels_r;                                                          \
2653   vec_8x16u pixels_g;                                                          \
2654   vec_8x16u pixels_b;                                                          \
2655                                                                                \
2656   vec_8x16u draw_mask;                                                         \
2657   vec_8x16u zero_mask;                                                         \
2658                                                                                \
2659   vec_8x8u d64_0x07;                                                           \
2660   vec_8x8u d64_0x1F;                                                           \
2661   vec_8x8u d64_1;                                                              \
2662   vec_8x8u d64_4;                                                              \
2663   vec_8x8u d64_128;                                                            \
2664                                                                                \
2665   vec_8x16u d128_0x8000;                                                       \
2666                                                                                \
2667   vec_8x16u test_mask;                                                         \
2668   u32 draw_mask_bits;                                                          \
2669                                                                                \
2670   gvld1q_u16(test_mask, psx_gpu->test_mask.e);                                 \
2671   shade_blocks_load_msb_mask_##target();                                       \
2672                                                                                \
2673   gvdup_n_u8(d64_0x07, 0x07);                                                  \
2674   gvdup_n_u8(d64_0x1F, 0x1F);                                                  \
2675   gvdup_n_u8(d64_1, 1);                                                        \
2676   gvdup_n_u8(d64_4, 4);                                                        \
2677   gvdup_n_u8(d64_128, 128u);                                                   \
2678                                                                                \
2679   gvdupq_n_u16(d128_0x8000, 0x8000);                                           \
2680                                                                                \
2681   shade_blocks_textured_modulated_##shading##_primitive_load(dithering,        \
2682    target);                                                                    \
2683                                                                                \
2684   while(num_blocks)                                                            \
2685   {                                                                            \
2686     draw_mask_bits = block->draw_mask_bits;                                    \
2687     gvdupq_n_u16(draw_mask, draw_mask_bits);                                   \
2688     gvtstq_u16(draw_mask, draw_mask, test_mask);                               \
2689                                                                                \
2690     shade_blocks_textured_modulated_##shading##_block_load();                  \
2691                                                                                \
2692     gvld1q_u16(texels, block->texels.e);                                       \
2693                                                                                \
2694     gvmovn_u16(texels_r, texels);                                              \
2695     gvshrn_n_u16(texels_g, texels, 5);                                         \
2696     gvshrn_n_u16(texels_b, texels, 7);                                         \
2697                                                                                \
2698     gvand(texels_r, texels_r, d64_0x1F);                                       \
2699     gvand(texels_g, texels_g, d64_0x1F);                                       \
2700     gvshr_n_u8(texels_b, texels_b, 3);                                         \
2701                                                                                \
2702     shade_blocks_textured_modulate_##dithering(r);                             \
2703     shade_blocks_textured_modulate_##dithering(g);                             \
2704     shade_blocks_textured_modulate_##dithering(b);                             \
2705                                                                                \
2706     gvceqzq_u16(zero_mask, texels);                                            \
2707     gvand(pixels, texels, d128_0x8000);                                        \
2708                                                                                \
2709     gvqshrun_n_s16(pixels_r_low, pixels_r, 4);                                 \
2710     gvqshrun_n_s16(pixels_g_low, pixels_g, 4);                                 \
2711     gvqshrun_n_s16(pixels_b_low, pixels_b, 4);                                 \
2712                                                                                \
2713     gvorrq(zero_mask, draw_mask, zero_mask);                                   \
2714                                                                                \
2715     gvshr_n_u8(pixels_r_low, pixels_r_low, 3);                                 \
2716     gvbic(pixels_g_low, pixels_g_low, d64_0x07);                               \
2717     gvbic(pixels_b_low, pixels_b_low, d64_0x07);                               \
2718                                                                                \
2719     gvmlal_u8(pixels, pixels_r_low, d64_1);                                    \
2720     gvmlal_u8(pixels, pixels_g_low, d64_4);                                    \
2721     gvmlal_u8(pixels, pixels_b_low, d64_128);                                  \
2722                                                                                \
2723     shade_blocks_store_##target(zero_mask, pixels);                            \
2724                                                                                \
2725     num_blocks--;                                                              \
2726     block++;                                                                   \
2727   }                                                                            \
2728
2729 void shade_blocks_shaded_textured_modulated_dithered_direct(psx_gpu_struct
2730  *psx_gpu)
2731 {
2732 #if 0
2733   shade_blocks_shaded_textured_modulated_dithered_direct_(psx_gpu);
2734   return;
2735 #endif
2736   shade_blocks_textured_modulated_do(shaded, dithered, direct);
2737 }
2738
2739 void shade_blocks_shaded_textured_modulated_undithered_direct(psx_gpu_struct
2740  *psx_gpu)
2741 {
2742 #if 0
2743   shade_blocks_shaded_textured_modulated_undithered_direct_(psx_gpu);
2744   return;
2745 #endif
2746   shade_blocks_textured_modulated_do(shaded, undithered, direct);
2747 }
2748
2749 void shade_blocks_unshaded_textured_modulated_dithered_direct(psx_gpu_struct
2750  *psx_gpu)
2751 {
2752 #if 0
2753   shade_blocks_unshaded_textured_modulated_dithered_direct_(psx_gpu);
2754   return;
2755 #endif
2756   shade_blocks_textured_modulated_do(unshaded, dithered, direct);
2757 }
2758
2759 void shade_blocks_unshaded_textured_modulated_undithered_direct(psx_gpu_struct
2760  *psx_gpu)
2761 {
2762 #if 0
2763   shade_blocks_unshaded_textured_modulated_undithered_direct_(psx_gpu);
2764   return;
2765 #endif
2766   shade_blocks_textured_modulated_do(unshaded, undithered, direct);
2767 }
2768
2769 void shade_blocks_shaded_textured_modulated_dithered_indirect(psx_gpu_struct
2770  *psx_gpu)
2771 {
2772 #if 0
2773   shade_blocks_shaded_textured_modulated_dithered_indirect_(psx_gpu);
2774   return;
2775 #endif
2776   shade_blocks_textured_modulated_do(shaded, dithered, indirect);
2777 }
2778
2779 void shade_blocks_shaded_textured_modulated_undithered_indirect(psx_gpu_struct
2780  *psx_gpu)
2781 {
2782 #if 0
2783   shade_blocks_shaded_textured_modulated_undithered_indirect_(psx_gpu);
2784   return;
2785 #endif
2786   shade_blocks_textured_modulated_do(shaded, undithered, indirect);
2787 }
2788
2789 void shade_blocks_unshaded_textured_modulated_dithered_indirect(psx_gpu_struct
2790  *psx_gpu)
2791 {
2792 #if 0
2793   shade_blocks_unshaded_textured_modulated_dithered_indirect_(psx_gpu);
2794   return;
2795 #endif
2796   shade_blocks_textured_modulated_do(unshaded, dithered, indirect);
2797 }
2798
2799 void shade_blocks_unshaded_textured_modulated_undithered_indirect(psx_gpu_struct
2800  *psx_gpu)
2801 {
2802 #if 0
2803   shade_blocks_unshaded_textured_modulated_undithered_indirect_(psx_gpu);
2804   return;
2805 #endif
2806   shade_blocks_textured_modulated_do(unshaded, undithered, indirect);
2807 }
2808
2809 #define shade_blocks_textured_unmodulated_do(target)                           \
2810   block_struct *block = psx_gpu->blocks;                                       \
2811   u32 num_blocks = psx_gpu->num_blocks;                                        \
2812   vec_8x16u draw_mask;                                                         \
2813   vec_8x16u test_mask;                                                         \
2814   u32 draw_mask_bits;                                                          \
2815                                                                                \
2816   vec_8x16u pixels;                                                            \
2817                                                                                \
2818   gvld1q_u16(test_mask, psx_gpu->test_mask.e);                                 \
2819   shade_blocks_load_msb_mask_##target();                                       \
2820                                                                                \
2821   while(num_blocks)                                                            \
2822   {                                                                            \
2823     vec_8x16u zero_mask;                                                       \
2824                                                                                \
2825     draw_mask_bits = block->draw_mask_bits;                                    \
2826     gvdupq_n_u16(draw_mask, draw_mask_bits);                                   \
2827     gvtstq_u16(draw_mask, draw_mask, test_mask);                               \
2828                                                                                \
2829     gvld1q_u16(pixels, block->texels.e);                                       \
2830                                                                                \
2831     gvceqzq_u16(zero_mask, pixels);                                            \
2832     gvorrq(zero_mask, draw_mask, zero_mask);                                   \
2833                                                                                \
2834     shade_blocks_store_##target(zero_mask, pixels);                            \
2835                                                                                \
2836     num_blocks--;                                                              \
2837     block++;                                                                   \
2838   }                                                                            \
2839
2840 void shade_blocks_textured_unmodulated_indirect(psx_gpu_struct *psx_gpu)
2841 {
2842 #if 0
2843   shade_blocks_textured_unmodulated_indirect_(psx_gpu);
2844   return;
2845 #endif
2846   shade_blocks_textured_unmodulated_do(indirect)
2847 }
2848
2849 void shade_blocks_textured_unmodulated_direct(psx_gpu_struct *psx_gpu)
2850 {
2851 #if 0
2852   shade_blocks_textured_unmodulated_direct_(psx_gpu);
2853   return;
2854 #endif
2855   shade_blocks_textured_unmodulated_do(direct)
2856 }
2857
2858 void shade_blocks_unshaded_untextured_indirect(psx_gpu_struct *psx_gpu)
2859 {
2860 }
2861
2862 void shade_blocks_unshaded_untextured_direct(psx_gpu_struct *psx_gpu)
2863 {
2864 #if 0
2865   shade_blocks_unshaded_untextured_direct_(psx_gpu);
2866   return;
2867 #endif
2868   block_struct *block = psx_gpu->blocks;
2869   u32 num_blocks = psx_gpu->num_blocks;
2870
2871   vec_8x16u pixels;
2872   gvld1q_u16(pixels, block->pixels.e);
2873   shade_blocks_load_msb_mask_direct();
2874
2875   while(num_blocks)
2876   {
2877     vec_8x16u draw_mask;
2878     gvld1q_u16(draw_mask, block->draw_mask.e);
2879     shade_blocks_store_direct(draw_mask, pixels);
2880
2881     num_blocks--;
2882     block++;
2883   }
2884 }
2885
2886 #define blend_blocks_mask_evaluate_on()                                        \
2887   vec_8x16u mask_pixels;                                                       \
2888   gvcltzq_s16(mask_pixels, framebuffer_pixels);                                \
2889   gvorrq(draw_mask, draw_mask, mask_pixels)                                    \
2890
2891 #define blend_blocks_mask_evaluate_off()                                       \
2892
2893 #define blend_blocks_average()                                                 \
2894 {                                                                              \
2895   vec_8x16u pixels_no_msb;                                                     \
2896   vec_8x16u fb_pixels_no_msb;                                                  \
2897                                                                                \
2898   vec_8x16u d128_0x0421;                                                       \
2899                                                                                \
2900   gvdupq_n_u16(d128_0x0421, 0x0421);                                           \
2901                                                                                \
2902   gveorq(blend_pixels, pixels, framebuffer_pixels);                            \
2903   gvbicq(pixels_no_msb, pixels, d128_0x8000);                                  \
2904   gvand(blend_pixels, blend_pixels, d128_0x0421);                              \
2905   gvsubq_u16(blend_pixels, pixels_no_msb, blend_pixels);                       \
2906   gvbicq(fb_pixels_no_msb, framebuffer_pixels, d128_0x8000);                   \
2907   gvhaddq_u16(blend_pixels, fb_pixels_no_msb, blend_pixels);                   \
2908 }                                                                              \
2909
2910 #define blend_blocks_add()                                                     \
2911 {                                                                              \
2912   vec_8x16u pixels_rb, pixels_g;                                               \
2913   vec_8x16u fb_rb, fb_g;                                                       \
2914                                                                                \
2915   vec_8x16u d128_0x7C1F;                                                       \
2916   vec_8x16u d128_0x03E0;                                                       \
2917                                                                                \
2918   gvdupq_n_u16(d128_0x7C1F, 0x7C1F);                                           \
2919   gvdupq_n_u16(d128_0x03E0, 0x03E0);                                           \
2920                                                                                \
2921   gvand(pixels_rb, pixels, d128_0x7C1F);                                       \
2922   gvand(pixels_g, pixels, d128_0x03E0);                                        \
2923                                                                                \
2924   gvand(fb_rb, framebuffer_pixels, d128_0x7C1F);                               \
2925   gvand(fb_g, framebuffer_pixels, d128_0x03E0);                                \
2926                                                                                \
2927   gvaddq_u16(fb_rb, fb_rb, pixels_rb);                                         \
2928   gvaddq_u16(fb_g, fb_g, pixels_g);                                            \
2929                                                                                \
2930   gvminq_u8(fb_rb, fb_rb, d128_0x7C1F);                                        \
2931   gvminq_u16(fb_g, fb_g, d128_0x03E0);                                         \
2932                                                                                \
2933   gvorrq(blend_pixels, fb_rb, fb_g);                                           \
2934 }                                                                              \
2935
2936 #define blend_blocks_subtract()                                                \
2937 {                                                                              \
2938   vec_8x16u pixels_rb, pixels_g;                                               \
2939   vec_8x16u fb_rb, fb_g;                                                       \
2940                                                                                \
2941   vec_8x16u d128_0x7C1F;                                                       \
2942   vec_8x16u d128_0x03E0;                                                       \
2943                                                                                \
2944   gvdupq_n_u16(d128_0x7C1F, 0x7C1F);                                           \
2945   gvdupq_n_u16(d128_0x03E0, 0x03E0);                                           \
2946                                                                                \
2947   gvand(pixels_rb, pixels, d128_0x7C1F);                                       \
2948   gvand(pixels_g, pixels, d128_0x03E0);                                        \
2949                                                                                \
2950   gvand(fb_rb, framebuffer_pixels, d128_0x7C1F);                               \
2951   gvand(fb_g, framebuffer_pixels, d128_0x03E0);                                \
2952                                                                                \
2953   gvqsubq_u8(fb_rb, fb_rb, pixels_rb);                                         \
2954   gvqsubq_u16(fb_g, fb_g, pixels_g);                                           \
2955                                                                                \
2956   gvorrq(blend_pixels, fb_rb, fb_g);                                           \
2957 }                                                                              \
2958
2959 #define blend_blocks_add_fourth()                                              \
2960 {                                                                              \
2961   vec_8x16u pixels_rb, pixels_g;                                               \
2962   vec_8x16u pixels_fourth;                                                     \
2963   vec_8x16u fb_rb, fb_g;                                                       \
2964                                                                                \
2965   vec_8x16u d128_0x7C1F;                                                       \
2966   vec_8x16u d128_0x1C07;                                                       \
2967   vec_8x16u d128_0x03E0;                                                       \
2968   vec_8x16u d128_0x00E0;                                                       \
2969                                                                                \
2970   gvdupq_n_u16(d128_0x7C1F, 0x7C1F);                                           \
2971   gvdupq_n_u16(d128_0x1C07, 0x1C07);                                           \
2972   gvdupq_n_u16(d128_0x03E0, 0x03E0);                                           \
2973   gvdupq_n_u16(d128_0x00E0, 0x00E0);                                           \
2974                                                                                \
2975   gvshrq_n_u16(pixels_fourth, pixels, 2);                                      \
2976                                                                                \
2977   gvand(fb_rb, framebuffer_pixels, d128_0x7C1F);                               \
2978   gvand(fb_g, framebuffer_pixels, d128_0x03E0);                                \
2979                                                                                \
2980   gvand(pixels_rb, pixels_fourth, d128_0x1C07);                                \
2981   gvand(pixels_g, pixels_fourth, d128_0x00E0);                                 \
2982                                                                                \
2983   gvaddq_u16(fb_rb, fb_rb, pixels_rb);                                         \
2984   gvaddq_u16(fb_g, fb_g, pixels_g);                                            \
2985                                                                                \
2986   gvminq_u8(fb_rb, fb_rb, d128_0x7C1F);                                        \
2987   gvminq_u16(fb_g, fb_g, d128_0x03E0);                                         \
2988                                                                                \
2989   gvorrq(blend_pixels, fb_rb, fb_g);                                           \
2990 }                                                                              \
2991
2992 #define blend_blocks_blended_combine_textured()                                \
2993 {                                                                              \
2994   vec_8x16u blend_mask;                                                        \
2995   gvcltzq_s16(blend_mask, pixels);                                             \
2996                                                                                \
2997   gvorrq(blend_pixels, blend_pixels, d128_0x8000);                             \
2998   gvbifq(blend_pixels, pixels, blend_mask);                                    \
2999 }                                                                              \
3000
3001 #define blend_blocks_blended_combine_untextured()                              \
3002
3003 #define blend_blocks_body_blend(blend_mode, texturing)                         \
3004 {                                                                              \
3005   blend_blocks_##blend_mode();                                                 \
3006   blend_blocks_blended_combine_##texturing();                                  \
3007 }                                                                              \
3008
3009 #define blend_blocks_body_average(texturing)                                   \
3010   blend_blocks_body_blend(average, texturing)                                  \
3011
3012 #define blend_blocks_body_add(texturing)                                       \
3013   blend_blocks_body_blend(add, texturing)                                      \
3014
3015 #define blend_blocks_body_subtract(texturing)                                  \
3016   blend_blocks_body_blend(subtract, texturing)                                 \
3017
3018 #define blend_blocks_body_add_fourth(texturing)                                \
3019   blend_blocks_body_blend(add_fourth, texturing)                               \
3020
3021 #define blend_blocks_body_unblended(texturing)                                 \
3022   blend_pixels = pixels                                                        \
3023
3024 #define blend_blocks_do(texturing, blend_mode, mask_evaluate)                  \
3025   block_struct *block = psx_gpu->blocks;                                       \
3026   u32 num_blocks = psx_gpu->num_blocks;                                        \
3027   vec_8x16u draw_mask;                                                         \
3028   vec_8x16u pixels;                                                            \
3029   vec_8x16u blend_pixels;                                                      \
3030   vec_8x16u framebuffer_pixels;                                                \
3031   vec_8x16u msb_mask;                                                          \
3032   vec_8x16u d128_0x8000;                                                       \
3033                                                                                \
3034   u16 *fb_ptr;                                                                 \
3035                                                                                \
3036   gvdupq_n_u16(d128_0x8000, 0x8000);                                           \
3037   gvdupq_n_u16(msb_mask, psx_gpu->mask_msb);                                   \
3038   (void)d128_0x8000; /* sometimes unused */                                    \
3039                                                                                \
3040   while(num_blocks)                                                            \
3041   {                                                                            \
3042     gvld1q_u16(pixels, block->pixels.e);                                       \
3043     gvld1q_u16(draw_mask, block->draw_mask.e);                                 \
3044     fb_ptr = block->fb_ptr;                                                    \
3045                                                                                \
3046     gvld1q_u16(framebuffer_pixels, fb_ptr);                                    \
3047                                                                                \
3048     blend_blocks_mask_evaluate_##mask_evaluate();                              \
3049     blend_blocks_body_##blend_mode(texturing);                                 \
3050                                                                                \
3051     gvorrq(blend_pixels, blend_pixels, msb_mask);                              \
3052     gvbifq(framebuffer_pixels, blend_pixels, draw_mask);                       \
3053     gvst1q_u16(framebuffer_pixels, fb_ptr);                                    \
3054                                                                                \
3055     num_blocks--;                                                              \
3056     block++;                                                                   \
3057   }                                                                            \
3058
3059
3060 void blend_blocks_textured_average_off(psx_gpu_struct *psx_gpu)
3061 {
3062 #if 0
3063   blend_blocks_textured_average_off_(psx_gpu);
3064   return;
3065 #endif
3066   blend_blocks_do(textured, average, off);
3067 }
3068
3069 void blend_blocks_untextured_average_off(psx_gpu_struct *psx_gpu)
3070 {
3071 #if 0
3072   blend_blocks_untextured_average_off_(psx_gpu);
3073   return;
3074 #endif
3075   blend_blocks_do(untextured, average, off);
3076 }
3077
3078 void blend_blocks_textured_average_on(psx_gpu_struct *psx_gpu)
3079 {
3080 #if 0
3081   blend_blocks_textured_average_on_(psx_gpu);
3082   return;
3083 #endif
3084   blend_blocks_do(textured, average, on);
3085 }
3086
3087 void blend_blocks_untextured_average_on(psx_gpu_struct *psx_gpu)
3088 {
3089 #if 0
3090   blend_blocks_untextured_average_on_(psx_gpu);
3091   return;
3092 #endif
3093   blend_blocks_do(untextured, average, on);
3094 }
3095
3096 void blend_blocks_textured_add_off(psx_gpu_struct *psx_gpu)
3097 {
3098 #if 0
3099   blend_blocks_textured_add_off_(psx_gpu);
3100   return;
3101 #endif
3102   blend_blocks_do(textured, add, off);
3103 }
3104
3105 void blend_blocks_textured_add_on(psx_gpu_struct *psx_gpu)
3106 {
3107 #if 0
3108   blend_blocks_textured_add_on_(psx_gpu);
3109   return;
3110 #endif
3111   blend_blocks_do(textured, add, on);
3112 }
3113
3114 void blend_blocks_untextured_add_off(psx_gpu_struct *psx_gpu)
3115 {
3116 #if 0
3117   blend_blocks_untextured_add_off_(psx_gpu);
3118   return;
3119 #endif
3120   blend_blocks_do(untextured, add, off);
3121 }
3122
3123 void blend_blocks_untextured_add_on(psx_gpu_struct *psx_gpu)
3124 {
3125 #if 0
3126   blend_blocks_untextured_add_on_(psx_gpu);
3127   return;
3128 #endif
3129   blend_blocks_do(untextured, add, on);
3130 }
3131
3132 void blend_blocks_textured_subtract_off(psx_gpu_struct *psx_gpu)
3133 {
3134 #if 0
3135   blend_blocks_textured_subtract_off_(psx_gpu);
3136   return;
3137 #endif
3138   blend_blocks_do(textured, subtract, off);
3139 }
3140
3141 void blend_blocks_textured_subtract_on(psx_gpu_struct *psx_gpu)
3142 {
3143 #if 0
3144   blend_blocks_textured_subtract_on_(psx_gpu);
3145   return;
3146 #endif
3147   blend_blocks_do(textured, subtract, on);
3148 }
3149
3150 void blend_blocks_untextured_subtract_off(psx_gpu_struct *psx_gpu)
3151 {
3152 #if 0
3153   blend_blocks_untextured_subtract_off_(psx_gpu);
3154   return;
3155 #endif
3156   blend_blocks_do(untextured, subtract, off);
3157 }
3158
3159 void blend_blocks_untextured_subtract_on(psx_gpu_struct *psx_gpu)
3160 {
3161 #if 0
3162   blend_blocks_untextured_subtract_on_(psx_gpu);
3163   return;
3164 #endif
3165   blend_blocks_do(untextured, subtract, on);
3166 }
3167
3168 void blend_blocks_textured_add_fourth_off(psx_gpu_struct *psx_gpu)
3169 {
3170 #if 0
3171   blend_blocks_textured_add_fourth_off_(psx_gpu);
3172   return;
3173 #endif
3174   blend_blocks_do(textured, add_fourth, off);
3175 }
3176
3177 void blend_blocks_textured_add_fourth_on(psx_gpu_struct *psx_gpu)
3178 {
3179 #if 0
3180   blend_blocks_textured_add_fourth_on_(psx_gpu);
3181   return;
3182 #endif
3183   blend_blocks_do(textured, add_fourth, on);
3184 }
3185
3186 void blend_blocks_untextured_add_fourth_off(psx_gpu_struct *psx_gpu)
3187 {
3188 #if 0
3189   blend_blocks_untextured_add_fourth_off_(psx_gpu);
3190   return;
3191 #endif
3192   blend_blocks_do(untextured, add_fourth, off);
3193 }
3194
3195 void blend_blocks_untextured_add_fourth_on(psx_gpu_struct *psx_gpu)
3196 {
3197 #if 0
3198   blend_blocks_untextured_add_fourth_on_(psx_gpu);
3199   return;
3200 #endif
3201   blend_blocks_do(untextured, add_fourth, on);
3202 }
3203
3204 void blend_blocks_textured_unblended_on(psx_gpu_struct *psx_gpu)
3205 {
3206 #if 0
3207   blend_blocks_textured_unblended_on_(psx_gpu);
3208   return;
3209 #endif
3210   blend_blocks_do(textured, unblended, on);
3211 }
3212
3213 void blend_blocks_textured_unblended_off(psx_gpu_struct *psx_gpu)
3214 {
3215 }
3216
3217 void setup_sprite_untextured_512(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u,
3218  s32 v, s32 width, s32 height, u32 color)
3219 {
3220 #if 0
3221   setup_sprite_untextured_512_(psx_gpu, x, y, u, v, width, height, color);
3222   return;
3223 #endif
3224   u32 right_width = ((width - 1) & 0x7) + 1;
3225   u32 right_mask_bits = (0xFF << right_width);
3226   u16 *fb_ptr = psx_gpu->vram_out_ptr + (y * 1024) + x;
3227   u32 block_width = (width + 7) / 8;
3228   u32 fb_ptr_pitch = 1024 - ((block_width - 1) * 8);
3229   u32 blocks_remaining;
3230   u32 num_blocks = psx_gpu->num_blocks;
3231   block_struct *block = psx_gpu->blocks + num_blocks;
3232
3233   u32 color_r = color & 0xFF;
3234   u32 color_g = (color >> 8) & 0xFF;
3235   u32 color_b = (color >> 16) & 0xFF;
3236   vec_8x16u colors;
3237   vec_8x16u right_mask;
3238   vec_8x16u test_mask;
3239   vec_8x16u zero_mask;
3240
3241   gvld1q_u16(test_mask, psx_gpu->test_mask.e);
3242   color = (color_r >> 3) | ((color_g >> 3) << 5) | ((color_b >> 3) << 10);
3243
3244   gvdupq_n_u16(colors, color);
3245   gvdupq_n_u16(zero_mask, 0x00);
3246   gvdupq_n_u16(right_mask, right_mask_bits);
3247   gvtstq_u16(right_mask, right_mask, test_mask);
3248
3249   while(height)
3250   {
3251     blocks_remaining = block_width - 1;
3252     num_blocks += block_width;
3253
3254     if(num_blocks > MAX_BLOCKS)
3255     {
3256       flush_render_block_buffer(psx_gpu);
3257       num_blocks = block_width;
3258       block = psx_gpu->blocks;
3259     }
3260
3261     while(blocks_remaining)
3262     {
3263       gvst1q_u16(colors, block->pixels.e);
3264       gvst1q_u16(zero_mask, block->draw_mask.e);
3265       block->fb_ptr = fb_ptr;
3266
3267       fb_ptr += 8;
3268       block++;
3269       blocks_remaining--;
3270     }
3271
3272     gvst1q_u16(colors, block->pixels.e);
3273     gvst1q_u16(right_mask, block->draw_mask.e);
3274     block->fb_ptr = fb_ptr;
3275
3276     block++;
3277     fb_ptr += fb_ptr_pitch;
3278
3279     height--;
3280     psx_gpu->num_blocks = num_blocks;
3281   }
3282 }
3283
3284 #define setup_sprite_tiled_initialize_4bpp_clut()                              \
3285   vec_16x8u clut_low, clut_high;                                               \
3286                                                                                \
3287   gvld2q_u8(clut_low, clut_high, (u8 *)psx_gpu->clut_ptr)                      \
3288
3289 #define setup_sprite_tiled_initialize_4bpp()                                   \
3290   setup_sprite_tiled_initialize_4bpp_clut();                                   \
3291                                                                                \
3292   if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_4bpp_mask)        \
3293     update_texture_4bpp_cache(psx_gpu)                                         \
3294
3295 #define setup_sprite_tiled_initialize_8bpp()                                   \
3296   if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_8bpp_mask)        \
3297     update_texture_8bpp_cache(psx_gpu)                                         \
3298
3299 #define setup_sprite_tile_fetch_texel_block_8bpp(offset)                       \
3300   texture_block_ptr = psx_gpu->texture_page_ptr +                              \
3301    ((texture_offset + offset) & texture_mask);                                 \
3302                                                                                \
3303   gvld1_u8(texels, (u8 *)texture_block_ptr)                                    \
3304
3305 #define setup_sprite_tile_add_blocks(tile_num_blocks)                          \
3306   num_blocks += tile_num_blocks;                                               \
3307                                                                                \
3308   if(num_blocks > MAX_BLOCKS)                                                  \
3309   {                                                                            \
3310     flush_render_block_buffer(psx_gpu);                                        \
3311     num_blocks = tile_num_blocks;                                              \
3312     block = psx_gpu->blocks;                                                   \
3313   }                                                                            \
3314
3315 #define setup_sprite_tile_full_4bpp(edge)                                      \
3316 {                                                                              \
3317   vec_8x8u texels_low, texels_high;                                            \
3318   setup_sprite_tile_add_blocks(sub_tile_height * 2);                           \
3319                                                                                \
3320   while(sub_tile_height)                                                       \
3321   {                                                                            \
3322     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3323     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3324     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3325                                                                                \
3326     gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e);                  \
3327     block->draw_mask_bits = left_mask_bits;                                    \
3328     block->fb_ptr = fb_ptr;                                                    \
3329     block++;                                                                   \
3330                                                                                \
3331     setup_sprite_tile_fetch_texel_block_8bpp(8);                               \
3332     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3333     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3334                                                                                \
3335     gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e);                  \
3336     block->draw_mask_bits = right_mask_bits;                                   \
3337     block->fb_ptr = fb_ptr + 8;                                                \
3338     block++;                                                                   \
3339                                                                                \
3340     fb_ptr += 1024;                                                            \
3341     texture_offset += 0x10;                                                    \
3342     sub_tile_height--;                                                         \
3343   }                                                                            \
3344   texture_offset += 0xF00;                                                     \
3345   psx_gpu->num_blocks = num_blocks;                                            \
3346 }                                                                              \
3347
3348 #define setup_sprite_tile_half_4bpp(edge)                                      \
3349 {                                                                              \
3350   vec_8x8u texels_low, texels_high;                                            \
3351   setup_sprite_tile_add_blocks(sub_tile_height);                               \
3352                                                                                \
3353   while(sub_tile_height)                                                       \
3354   {                                                                            \
3355     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3356     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3357     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3358                                                                                \
3359     gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e);                  \
3360     block->draw_mask_bits = edge##_mask_bits;                                  \
3361     block->fb_ptr = fb_ptr;                                                    \
3362     block++;                                                                   \
3363                                                                                \
3364     fb_ptr += 1024;                                                            \
3365     texture_offset += 0x10;                                                    \
3366     sub_tile_height--;                                                         \
3367   }                                                                            \
3368   texture_offset += 0xF00;                                                     \
3369   psx_gpu->num_blocks = num_blocks;                                            \
3370 }                                                                              \
3371
3372 #define setup_sprite_tile_full_8bpp(edge)                                      \
3373 {                                                                              \
3374   setup_sprite_tile_add_blocks(sub_tile_height * 2);                           \
3375                                                                                \
3376   while(sub_tile_height)                                                       \
3377   {                                                                            \
3378     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3379     gvst1_u8(texels, block->r.e);                                              \
3380     block->draw_mask_bits = left_mask_bits;                                    \
3381     block->fb_ptr = fb_ptr;                                                    \
3382     block++;                                                                   \
3383                                                                                \
3384     setup_sprite_tile_fetch_texel_block_8bpp(8);                               \
3385     gvst1_u8(texels, block->r.e);                                              \
3386     block->draw_mask_bits = right_mask_bits;                                   \
3387     block->fb_ptr = fb_ptr + 8;                                                \
3388     block++;                                                                   \
3389                                                                                \
3390     fb_ptr += 1024;                                                            \
3391     texture_offset += 0x10;                                                    \
3392     sub_tile_height--;                                                         \
3393   }                                                                            \
3394   texture_offset += 0xF00;                                                     \
3395   psx_gpu->num_blocks = num_blocks;                                            \
3396 }                                                                              \
3397
3398 #define setup_sprite_tile_half_8bpp(edge)                                      \
3399 {                                                                              \
3400   setup_sprite_tile_add_blocks(sub_tile_height);                               \
3401                                                                                \
3402   while(sub_tile_height)                                                       \
3403   {                                                                            \
3404     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3405     gvst1_u8(texels, block->r.e);                                              \
3406     block->draw_mask_bits = edge##_mask_bits;                                  \
3407     block->fb_ptr = fb_ptr;                                                    \
3408     block++;                                                                   \
3409                                                                                \
3410     fb_ptr += 1024;                                                            \
3411     texture_offset += 0x10;                                                    \
3412     sub_tile_height--;                                                         \
3413   }                                                                            \
3414   texture_offset += 0xF00;                                                     \
3415   psx_gpu->num_blocks = num_blocks;                                            \
3416 }                                                                              \
3417
3418 #define setup_sprite_tile_column_edge_pre_adjust_half_right()                  \
3419   texture_offset = texture_offset_base + 8;                                    \
3420   fb_ptr += 8                                                                  \
3421
3422 #define setup_sprite_tile_column_edge_pre_adjust_half_left()                   \
3423   texture_offset = texture_offset_base                                         \
3424
3425 #define setup_sprite_tile_column_edge_pre_adjust_half(edge)                    \
3426   setup_sprite_tile_column_edge_pre_adjust_half_##edge()                       \
3427
3428 #define setup_sprite_tile_column_edge_pre_adjust_full(edge)                    \
3429   texture_offset = texture_offset_base                                         \
3430
3431 #define setup_sprite_tile_column_edge_post_adjust_half_right()                 \
3432   fb_ptr -= 8                                                                  \
3433
3434 #define setup_sprite_tile_column_edge_post_adjust_half_left()                  \
3435
3436 #define setup_sprite_tile_column_edge_post_adjust_half(edge)                   \
3437   setup_sprite_tile_column_edge_post_adjust_half_##edge()                      \
3438
3439 #define setup_sprite_tile_column_edge_post_adjust_full(edge)                   \
3440
3441
3442 #define setup_sprite_tile_column_height_single(edge_mode, edge, texture_mode,  \
3443  x4mode)                                                                       \
3444 do                                                                             \
3445 {                                                                              \
3446   sub_tile_height = column_data;                                               \
3447   setup_sprite_tile_column_edge_pre_adjust_##edge_mode##x4mode(edge);          \
3448   setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge);                \
3449   setup_sprite_tile_column_edge_post_adjust_##edge_mode##x4mode(edge);         \
3450 } while(0)                                                                     \
3451
3452 #define setup_sprite_tile_column_height_multi(edge_mode, edge, texture_mode,   \
3453  x4mode)                                                                       \
3454 do                                                                             \
3455 {                                                                              \
3456   u32 tiles_remaining = column_data >> 16;                                     \
3457   sub_tile_height = column_data & 0xFF;                                        \
3458   setup_sprite_tile_column_edge_pre_adjust_##edge_mode##x4mode(edge);          \
3459   setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge);                \
3460   tiles_remaining -= 1;                                                        \
3461                                                                                \
3462   while(tiles_remaining)                                                       \
3463   {                                                                            \
3464     sub_tile_height = 16;                                                      \
3465     setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge);              \
3466     tiles_remaining--;                                                         \
3467   }                                                                            \
3468                                                                                \
3469   sub_tile_height = (column_data >> 8) & 0xFF;                                 \
3470   setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge);                \
3471   setup_sprite_tile_column_edge_post_adjust_##edge_mode##x4mode(edge);         \
3472 } while(0)                                                                     \
3473
3474
3475 #define setup_sprite_column_data_single()                                      \
3476   column_data = height                                                         \
3477
3478 #define setup_sprite_column_data_multi()                                       \
3479   column_data = 16 - offset_v;                                                 \
3480   column_data |= ((height_rounded & 0xF) + 1) << 8;                            \
3481   column_data |= (tile_height - 1) << 16                                       \
3482
3483 #define RIGHT_MASK_BIT_SHIFT 8
3484 #define RIGHT_MASK_BIT_SHIFT_4x 16
3485
3486 #define setup_sprite_tile_column_width_single(texture_mode, multi_height,      \
3487  edge_mode, edge, x4mode)                                                      \
3488 {                                                                              \
3489   setup_sprite_column_data_##multi_height();                                   \
3490   left_mask_bits = left_block_mask | right_block_mask;                         \
3491   right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode;            \
3492                                                                                \
3493   setup_sprite_tile_column_height_##multi_height(edge_mode, edge,              \
3494    texture_mode, x4mode);                                                      \
3495 }                                                                              \
3496
3497 #define setup_sprite_tiled_advance_column()                                    \
3498   texture_offset_base += 0x100;                                                \
3499   if((texture_offset_base & 0xF00) == 0)                                       \
3500     texture_offset_base -= (0x100 + 0xF00)                                     \
3501
3502 #define FB_PTR_MULTIPLIER 1
3503 #define FB_PTR_MULTIPLIER_4x 2
3504
3505 #define setup_sprite_tile_column_width_multi(texture_mode, multi_height,       \
3506  left_mode, right_mode, x4mode)                                                \
3507 {                                                                              \
3508   setup_sprite_column_data_##multi_height();                                   \
3509   s32 fb_ptr_advance_column = (16 - (1024 * height))                           \
3510     * FB_PTR_MULTIPLIER##x4mode;                                               \
3511                                                                                \
3512   tile_width -= 2;                                                             \
3513   left_mask_bits = left_block_mask;                                            \
3514   right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode;            \
3515                                                                                \
3516   setup_sprite_tile_column_height_##multi_height(left_mode, right,             \
3517    texture_mode, x4mode);                                                      \
3518   fb_ptr += fb_ptr_advance_column;                                             \
3519                                                                                \
3520   left_mask_bits = 0x00;                                                       \
3521   right_mask_bits = 0x00;                                                      \
3522                                                                                \
3523   while(tile_width)                                                            \
3524   {                                                                            \
3525     setup_sprite_tiled_advance_column();                                       \
3526     setup_sprite_tile_column_height_##multi_height(full, none,                 \
3527      texture_mode, x4mode);                                                    \
3528     fb_ptr += fb_ptr_advance_column;                                           \
3529     tile_width--;                                                              \
3530   }                                                                            \
3531                                                                                \
3532   left_mask_bits = right_block_mask;                                           \
3533   right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode;            \
3534                                                                                \
3535   setup_sprite_tiled_advance_column();                                         \
3536   setup_sprite_tile_column_height_##multi_height(right_mode, left,             \
3537    texture_mode, x4mode);                                                      \
3538 }                                                                              \
3539
3540
3541 /* 4x stuff */
3542 #define setup_sprite_tiled_initialize_4bpp_4x()                                \
3543   setup_sprite_tiled_initialize_4bpp_clut()                                    \
3544
3545 #define setup_sprite_tiled_initialize_8bpp_4x()                                \
3546
3547 #define setup_sprite_tile_full_4bpp_4x(edge)                                   \
3548 {                                                                              \
3549   vec_8x8u texels_low, texels_high;                                            \
3550   vec_8x16u pixels;                                                            \
3551   vec_4x16u pixels_half;                                                       \
3552   setup_sprite_tile_add_blocks(sub_tile_height * 2 * 4);                       \
3553   u32 left_mask_bits_a = left_mask_bits & 0xFF;                                \
3554   u32 left_mask_bits_b = left_mask_bits >> 8;                                  \
3555   u32 right_mask_bits_a = right_mask_bits & 0xFF;                              \
3556   u32 right_mask_bits_b = right_mask_bits >> 8;                                \
3557                                                                                \
3558   while(sub_tile_height)                                                       \
3559   {                                                                            \
3560     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3561     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3562     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3563     gvzip_u8(pixels, texels_low, texels_high);                                 \
3564                                                                                \
3565     gvget_lo(pixels_half, pixels);                                             \
3566     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3567     block->draw_mask_bits = left_mask_bits_a;                                  \
3568     block->fb_ptr = fb_ptr;                                                    \
3569     block++;                                                                   \
3570                                                                                \
3571     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3572     block->draw_mask_bits = left_mask_bits_a;                                  \
3573     block->fb_ptr = fb_ptr + 1024;                                             \
3574     block++;                                                                   \
3575                                                                                \
3576     gvget_hi(pixels_half, pixels);                                             \
3577     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3578     block->draw_mask_bits = left_mask_bits_b;                                  \
3579     block->fb_ptr = fb_ptr + 8;                                                \
3580     block++;                                                                   \
3581                                                                                \
3582     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3583     block->draw_mask_bits = left_mask_bits_b;                                  \
3584     block->fb_ptr = fb_ptr + 1024 + 8;                                         \
3585     block++;                                                                   \
3586                                                                                \
3587     setup_sprite_tile_fetch_texel_block_8bpp(8);                               \
3588     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3589     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3590     gvzip_u8(pixels, texels_low, texels_high);                                 \
3591                                                                                \
3592     gvget_lo(pixels_half, pixels);                                             \
3593     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3594     block->draw_mask_bits = right_mask_bits_a;                                 \
3595     block->fb_ptr = fb_ptr + 16;                                               \
3596     block++;                                                                   \
3597                                                                                \
3598     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3599     block->draw_mask_bits = right_mask_bits_a;                                 \
3600     block->fb_ptr = fb_ptr + 1024 + 16;                                        \
3601     block++;                                                                   \
3602                                                                                \
3603     gvget_hi(pixels_half, pixels);                                             \
3604     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3605     block->draw_mask_bits = right_mask_bits_b;                                 \
3606     block->fb_ptr = fb_ptr + 24;                                               \
3607     block++;                                                                   \
3608                                                                                \
3609     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3610     block->draw_mask_bits = right_mask_bits_b;                                 \
3611     block->fb_ptr = fb_ptr + 1024 + 24;                                        \
3612     block++;                                                                   \
3613                                                                                \
3614     fb_ptr += 2048;                                                            \
3615     texture_offset += 0x10;                                                    \
3616     sub_tile_height--;                                                         \
3617   }                                                                            \
3618   texture_offset += 0xF00;                                                     \
3619   psx_gpu->num_blocks = num_blocks;                                            \
3620 }                                                                              \
3621
3622 #define setup_sprite_tile_half_4bpp_4x(edge)                                   \
3623 {                                                                              \
3624   vec_8x8u texels_low, texels_high;                                            \
3625   vec_8x16u pixels;                                                            \
3626   vec_4x16u pixels_half;                                                       \
3627   setup_sprite_tile_add_blocks(sub_tile_height * 4);                           \
3628   u32 edge##_mask_bits_a = edge##_mask_bits & 0xFF;                            \
3629   u32 edge##_mask_bits_b = edge##_mask_bits >> 8;                              \
3630                                                                                \
3631   while(sub_tile_height)                                                       \
3632   {                                                                            \
3633     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3634     gvtbl2_u8(texels_low, clut_low, texels);                                   \
3635     gvtbl2_u8(texels_high, clut_high, texels);                                 \
3636     gvzip_u8(pixels, texels_low, texels_high);                                 \
3637                                                                                \
3638     gvget_lo(pixels_half, pixels);                                             \
3639     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3640     block->draw_mask_bits = edge##_mask_bits_a;                                \
3641     block->fb_ptr = fb_ptr;                                                    \
3642     block++;                                                                   \
3643                                                                                \
3644     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3645     block->draw_mask_bits = edge##_mask_bits_a;                                \
3646     block->fb_ptr = fb_ptr + 1024;                                             \
3647     block++;                                                                   \
3648                                                                                \
3649     gvget_hi(pixels_half, pixels);                                             \
3650     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3651     block->draw_mask_bits = edge##_mask_bits_b;                                \
3652     block->fb_ptr = fb_ptr + 8;                                                \
3653     block++;                                                                   \
3654                                                                                \
3655     gvst2_u16(pixels_half, pixels_half, block->texels.e);                      \
3656     block->draw_mask_bits = edge##_mask_bits_b;                                \
3657     block->fb_ptr = fb_ptr + 1024 + 8;                                         \
3658     block++;                                                                   \
3659                                                                                \
3660     fb_ptr += 2048;                                                            \
3661     texture_offset += 0x10;                                                    \
3662     sub_tile_height--;                                                         \
3663   }                                                                            \
3664   texture_offset += 0xF00;                                                     \
3665   psx_gpu->num_blocks = num_blocks;                                            \
3666 }                                                                              \
3667
3668 #define setup_sprite_tile_full_8bpp_4x(edge)                                   \
3669 {                                                                              \
3670   setup_sprite_tile_add_blocks(sub_tile_height * 2 * 4);                       \
3671   vec_8x16u texels_wide;                                                       \
3672   vec_4x16u texels_half;                                                       \
3673   u32 left_mask_bits_a = left_mask_bits & 0xFF;                                \
3674   u32 left_mask_bits_b = left_mask_bits >> 8;                                  \
3675   u32 right_mask_bits_a = right_mask_bits & 0xFF;                              \
3676   u32 right_mask_bits_b = right_mask_bits >> 8;                                \
3677                                                                                \
3678   while(sub_tile_height)                                                       \
3679   {                                                                            \
3680     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3681     gvzip_u8(texels_wide, texels, texels);                                     \
3682     gvget_lo(texels_half, texels_wide);                                        \
3683     gvst1_u8(texels_half, block->r.e);                                         \
3684     block->draw_mask_bits = left_mask_bits_a;                                  \
3685     block->fb_ptr = fb_ptr;                                                    \
3686     block++;                                                                   \
3687                                                                                \
3688     gvst1_u8(texels_half, block->r.e);                                         \
3689     block->draw_mask_bits = left_mask_bits_a;                                  \
3690     block->fb_ptr = fb_ptr + 1024;                                             \
3691     block++;                                                                   \
3692                                                                                \
3693     gvget_hi(texels_half, texels_wide);                                        \
3694     gvst1_u8(texels_half, block->r.e);                                         \
3695     block->draw_mask_bits = left_mask_bits_b;                                  \
3696     block->fb_ptr = fb_ptr + 8;                                                \
3697     block++;                                                                   \
3698                                                                                \
3699     gvst1_u8(texels_half, block->r.e);                                         \
3700     block->draw_mask_bits = left_mask_bits_b;                                  \
3701     block->fb_ptr = fb_ptr + 1024 + 8;                                         \
3702     block++;                                                                   \
3703                                                                                \
3704     setup_sprite_tile_fetch_texel_block_8bpp(8);                               \
3705     gvzip_u8(texels_wide, texels, texels);                                     \
3706     gvget_lo(texels_half, texels_wide);                                        \
3707     gvst1_u8(texels_half, block->r.e);                                         \
3708     block->draw_mask_bits = right_mask_bits_a;                                 \
3709     block->fb_ptr = fb_ptr + 16;                                               \
3710     block++;                                                                   \
3711                                                                                \
3712     gvst1_u8(texels_half, block->r.e);                                         \
3713     block->draw_mask_bits = right_mask_bits_a;                                 \
3714     block->fb_ptr = fb_ptr + 1024 + 16;                                        \
3715     block++;                                                                   \
3716                                                                                \
3717     gvget_hi(texels_half, texels_wide);                                        \
3718     gvst1_u8(texels_half, block->r.e);                                         \
3719     block->draw_mask_bits = right_mask_bits_b;                                 \
3720     block->fb_ptr = fb_ptr + 24;                                               \
3721     block++;                                                                   \
3722                                                                                \
3723     gvst1_u8(texels_half, block->r.e);                                         \
3724     block->draw_mask_bits = right_mask_bits_b;                                 \
3725     block->fb_ptr = fb_ptr + 24 + 1024;                                        \
3726     block++;                                                                   \
3727                                                                                \
3728     fb_ptr += 2048;                                                            \
3729     texture_offset += 0x10;                                                    \
3730     sub_tile_height--;                                                         \
3731   }                                                                            \
3732   texture_offset += 0xF00;                                                     \
3733   psx_gpu->num_blocks = num_blocks;                                            \
3734 }                                                                              \
3735
3736 #define setup_sprite_tile_half_8bpp_4x(edge)                                   \
3737 {                                                                              \
3738   setup_sprite_tile_add_blocks(sub_tile_height * 4);                           \
3739   vec_8x16u texels_wide;                                                       \
3740   vec_4x16u texels_half;                                                       \
3741   u32 edge##_mask_bits_a = edge##_mask_bits & 0xFF;                            \
3742   u32 edge##_mask_bits_b = edge##_mask_bits >> 8;                              \
3743                                                                                \
3744   while(sub_tile_height)                                                       \
3745   {                                                                            \
3746     setup_sprite_tile_fetch_texel_block_8bpp(0);                               \
3747     gvzip_u8(texels_wide, texels, texels);                                     \
3748     gvget_lo(texels_half, texels_wide);                                        \
3749     gvst1_u8(texels_half, block->r.e);                                         \
3750     block->draw_mask_bits = edge##_mask_bits_a;                                \
3751     block->fb_ptr = fb_ptr;                                                    \
3752     block++;                                                                   \
3753                                                                                \
3754     gvst1_u8(texels_half, block->r.e);                                         \
3755     block->draw_mask_bits = edge##_mask_bits_a;                                \
3756     block->fb_ptr = fb_ptr + 1024;                                             \
3757     block++;                                                                   \
3758                                                                                \
3759     gvget_hi(texels_half, texels_wide);                                        \
3760     gvst1_u8(texels_half, block->r.e);                                         \
3761     block->draw_mask_bits = edge##_mask_bits_b;                                \
3762     block->fb_ptr = fb_ptr + 8;                                                \
3763     block++;                                                                   \
3764                                                                                \
3765     gvst1_u8(texels_half, block->r.e);                                         \
3766     block->draw_mask_bits = edge##_mask_bits_b;                                \
3767     block->fb_ptr = fb_ptr + 8 + 1024;                                         \
3768     block++;                                                                   \
3769                                                                                \
3770     fb_ptr += 2048;                                                            \
3771     texture_offset += 0x10;                                                    \
3772     sub_tile_height--;                                                         \
3773   }                                                                            \
3774   texture_offset += 0xF00;                                                     \
3775   psx_gpu->num_blocks = num_blocks;                                            \
3776 }                                                                              \
3777
3778 #define setup_sprite_tile_column_edge_pre_adjust_half_right_4x()               \
3779   texture_offset = texture_offset_base + 8;                                    \
3780   fb_ptr += 16                                                                 \
3781
3782 #define setup_sprite_tile_column_edge_pre_adjust_half_left_4x()                \
3783   texture_offset = texture_offset_base                                         \
3784
3785 #define setup_sprite_tile_column_edge_pre_adjust_half_4x(edge)                 \
3786   setup_sprite_tile_column_edge_pre_adjust_half_##edge##_4x()                  \
3787
3788 #define setup_sprite_tile_column_edge_pre_adjust_full_4x(edge)                 \
3789   texture_offset = texture_offset_base                                         \
3790
3791 #define setup_sprite_tile_column_edge_post_adjust_half_right_4x()              \
3792   fb_ptr -= 16                                                                 \
3793
3794 #define setup_sprite_tile_column_edge_post_adjust_half_left_4x()               \
3795
3796 #define setup_sprite_tile_column_edge_post_adjust_half_4x(edge)                \
3797   setup_sprite_tile_column_edge_post_adjust_half_##edge##_4x()                 \
3798
3799 #define setup_sprite_tile_column_edge_post_adjust_full_4x(edge)                \
3800
3801 #define setup_sprite_offset_u_adjust()                                         \
3802
3803 #define setup_sprite_comapre_left_block_mask()                                 \
3804   ((left_block_mask & 0xFF) == 0xFF)                                           \
3805
3806 #define setup_sprite_comapre_right_block_mask()                                \
3807   (((right_block_mask >> 8) & 0xFF) == 0xFF)                                   \
3808
3809 #define setup_sprite_offset_u_adjust_4x()                                      \
3810   offset_u *= 2;                                                               \
3811   offset_u_right = offset_u_right * 2 + 1                                      \
3812
3813 #define setup_sprite_comapre_left_block_mask_4x()                              \
3814   ((left_block_mask & 0xFFFF) == 0xFFFF)                                       \
3815
3816 #define setup_sprite_comapre_right_block_mask_4x()                             \
3817   (((right_block_mask >> 16) & 0xFFFF) == 0xFFFF)                              \
3818
3819
3820 #define setup_sprite_tiled_do(texture_mode, x4mode)                            \
3821   s32 offset_u = u & 0xF;                                                      \
3822   s32 offset_v = v & 0xF;                                                      \
3823                                                                                \
3824   s32 width_rounded = offset_u + width + 15;                                   \
3825   s32 height_rounded = offset_v + height + 15;                                 \
3826   s32 tile_height = height_rounded / 16;                                       \
3827   s32 tile_width = width_rounded / 16;                                         \
3828   u32 offset_u_right = width_rounded & 0xF;                                    \
3829                                                                                \
3830   setup_sprite_offset_u_adjust##x4mode();                                      \
3831                                                                                \
3832   u32 left_block_mask = ~(0xFFFFFFFF << offset_u);                             \
3833   u32 right_block_mask = 0xFFFFFFFE << offset_u_right;                         \
3834                                                                                \
3835   u32 left_mask_bits;                                                          \
3836   u32 right_mask_bits;                                                         \
3837                                                                                \
3838   u32 sub_tile_height;                                                         \
3839   u32 column_data;                                                             \
3840                                                                                \
3841   u32 texture_mask = (psx_gpu->texture_mask_width & 0xF) |                     \
3842    ((psx_gpu->texture_mask_height & 0xF) << 4) |                               \
3843    ((psx_gpu->texture_mask_width >> 4) << 8) |                                 \
3844    ((psx_gpu->texture_mask_height >> 4) << 12);                                \
3845   u32 texture_offset = ((v & 0xF) << 4) | ((u & 0xF0) << 4) |                  \
3846    ((v & 0xF0) << 8);                                                          \
3847   u32 texture_offset_base = texture_offset;                                    \
3848   u32 control_mask;                                                            \
3849                                                                                \
3850   u16 *fb_ptr = psx_gpu->vram_out_ptr + (y * 1024) + (x - offset_u);           \
3851   u32 num_blocks = psx_gpu->num_blocks;                                        \
3852   block_struct *block = psx_gpu->blocks + num_blocks;                          \
3853                                                                                \
3854   u16 *texture_block_ptr;                                                      \
3855   vec_8x8u texels;                                                             \
3856                                                                                \
3857   setup_sprite_tiled_initialize_##texture_mode##x4mode();                      \
3858                                                                                \
3859   control_mask = tile_width == 1;                                              \
3860   control_mask |= (tile_height == 1) << 1;                                     \
3861   control_mask |= setup_sprite_comapre_left_block_mask##x4mode() << 2;         \
3862   control_mask |= setup_sprite_comapre_right_block_mask##x4mode() << 3;        \
3863                                                                                \
3864   switch(control_mask)                                                         \
3865   {                                                                            \
3866     default:                                                                   \
3867     case 0x0:                                                                  \
3868       setup_sprite_tile_column_width_multi(texture_mode, multi, full, full,    \
3869        x4mode);                                                                \
3870       break;                                                                   \
3871                                                                                \
3872     case 0x1:                                                                  \
3873       setup_sprite_tile_column_width_single(texture_mode, multi, full, none,   \
3874        x4mode);                                                                \
3875       break;                                                                   \
3876                                                                                \
3877     case 0x2:                                                                  \
3878       setup_sprite_tile_column_width_multi(texture_mode, single, full, full,   \
3879        x4mode);                                                                \
3880       break;                                                                   \
3881                                                                                \
3882     case 0x3:                                                                  \
3883       setup_sprite_tile_column_width_single(texture_mode, single, full, none,  \
3884        x4mode);                                                                \
3885       break;                                                                   \
3886                                                                                \
3887     case 0x4:                                                                  \
3888       setup_sprite_tile_column_width_multi(texture_mode, multi, half, full,    \
3889        x4mode);                                                                \
3890       break;                                                                   \
3891                                                                                \
3892     case 0x5:                                                                  \
3893       setup_sprite_tile_column_width_single(texture_mode, multi, half, right,  \
3894        x4mode);                                                                \
3895       break;                                                                   \
3896                                                                                \
3897     case 0x6:                                                                  \
3898       setup_sprite_tile_column_width_multi(texture_mode, single, half, full,   \
3899        x4mode);                                                                \
3900       break;                                                                   \
3901                                                                                \
3902     case 0x7:                                                                  \
3903       setup_sprite_tile_column_width_single(texture_mode, single, half, right, \
3904        x4mode);                                                                \
3905       break;                                                                   \
3906                                                                                \
3907     case 0x8:                                                                  \
3908       setup_sprite_tile_column_width_multi(texture_mode, multi, full, half,    \
3909        x4mode);                                                                \
3910       break;                                                                   \
3911                                                                                \
3912     case 0x9:                                                                  \
3913       setup_sprite_tile_column_width_single(texture_mode, multi, half, left,   \
3914        x4mode);                                                                \
3915       break;                                                                   \
3916                                                                                \
3917     case 0xA:                                                                  \
3918       setup_sprite_tile_column_width_multi(texture_mode, single, full, half,   \
3919        x4mode);                                                                \
3920       break;                                                                   \
3921                                                                                \
3922     case 0xB:                                                                  \
3923       setup_sprite_tile_column_width_single(texture_mode, single, half, left,  \
3924        x4mode);                                                                \
3925       break;                                                                   \
3926                                                                                \
3927     case 0xC:                                                                  \
3928       setup_sprite_tile_column_width_multi(texture_mode, multi, half, half,    \
3929        x4mode);                                                                \
3930       break;                                                                   \
3931                                                                                \
3932     case 0xE:                                                                  \
3933       setup_sprite_tile_column_width_multi(texture_mode, single, half, half,   \
3934        x4mode);                                                                \
3935       break;                                                                   \
3936   }                                                                            \
3937
3938 void setup_sprite_4bpp(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
3939  s32 width, s32 height, u32 color)
3940 {
3941 #if 0
3942   setup_sprite_4bpp_(psx_gpu, x, y, u, v, width, height, color);
3943   return;
3944 #endif
3945   setup_sprite_tiled_do(4bpp,)
3946 }
3947
3948 void setup_sprite_8bpp(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
3949  s32 width, s32 height, u32 color)
3950 {
3951 #if 0
3952   setup_sprite_8bpp_(psx_gpu, x, y, u, v, width, height, color);
3953   return;
3954 #endif
3955   setup_sprite_tiled_do(8bpp,)
3956 }
3957
3958 #undef draw_mask_fb_ptr_left
3959 #undef draw_mask_fb_ptr_right
3960
3961 void setup_sprite_4bpp_4x(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
3962  s32 width, s32 height, u32 color)
3963 {
3964 #if 0
3965   setup_sprite_4bpp_4x_(psx_gpu, x, y, u, v, width, height, color);
3966   return;
3967 #endif
3968   setup_sprite_tiled_do(4bpp, _4x)
3969 }
3970
3971 void setup_sprite_8bpp_4x(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
3972  s32 width, s32 height, u32 color)
3973 {
3974 #if 0
3975   setup_sprite_8bpp_4x_(psx_gpu, x, y, u, v, width, height, color);
3976   return;
3977 #endif
3978   setup_sprite_tiled_do(8bpp, _4x)
3979 }
3980
3981
3982 void scale2x_tiles8(void * __restrict__ dst_, const void * __restrict__ src_, int w8, int h)
3983 {
3984 #if 0
3985   scale2x_tiles8_(dst_, src_, w8, h);
3986   return;
3987 #endif
3988   const u16 * __restrict__ src = src_;
3989   const u16 * __restrict__ src1;
3990   u16 * __restrict__ dst = dst_;
3991   u16 * __restrict__ dst1;
3992   gvreg a, b;
3993   int w;
3994   for (; h > 0; h--, src += 1024, dst += 1024*2)
3995   {
3996     src1 = src;
3997     dst1 = dst;
3998     for (w = w8; w > 0; w--, src1 += 8, dst1 += 8*2)
3999     {
4000       gvld1q_u16(a, src1);
4001       gvzipq_u16(a, b, a, a);
4002       gvst1q_u16(a, dst1);
4003       gvst1q_u16(b, dst1 + 8);
4004       gvst1q_u16(a, dst1 + 1024);
4005       gvst1q_u16(b, dst1 + 1024 + 8);
4006     }
4007   }
4008 }
4009
4010 // vim:ts=2:sw=2:expandtab