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