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