--- /dev/null
+/*
+ * Copyright (C) 2011 Gilead Kutnick "Exophase" <exophase@gmail.com>
+ * Copyright (C) 2022 GraÅžvydas Ignotas "notaz" <notasas@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ */
+
+#include <string.h>
+#include "psx_gpu.h"
+#include "psx_gpu_simd.h"
+//#define ASM_PROTOTYPES
+//#include "psx_gpu_simd.h"
+#ifndef SIMD_BUILD
+#error "please define SIMD_BUILD if you want this gpu_neon C simd implementation"
+#endif
+
+typedef u8 gvu8 __attribute__((vector_size(16)));
+typedef u16 gvu16 __attribute__((vector_size(16)));
+typedef u32 gvu32 __attribute__((vector_size(16)));
+typedef u64 gvu64 __attribute__((vector_size(16)));
+typedef s8 gvs8 __attribute__((vector_size(16)));
+typedef s16 gvs16 __attribute__((vector_size(16)));
+typedef s32 gvs32 __attribute__((vector_size(16)));
+typedef s64 gvs64 __attribute__((vector_size(16)));
+
+typedef u8 gvhu8 __attribute__((vector_size(8)));
+typedef u16 gvhu16 __attribute__((vector_size(8)));
+typedef u32 gvhu32 __attribute__((vector_size(8)));
+typedef u64 gvhu64 __attribute__((vector_size(8)));
+typedef s8 gvhs8 __attribute__((vector_size(8)));
+typedef s16 gvhs16 __attribute__((vector_size(8)));
+typedef s32 gvhs32 __attribute__((vector_size(8)));
+typedef s64 gvhs64 __attribute__((vector_size(8)));
+
+typedef union
+{
+ gvhu8 u8;
+ gvhu16 u16;
+ gvhu32 u32;
+ gvhu64 u64;
+ //u64 u64;
+ //uint64x1_t u64;
+ gvhs8 s8;
+ gvhs16 s16;
+ gvhs32 s32;
+ gvhs64 s64;
+ //s64 s64;
+ //int64x1_t s64;
+} gvhreg;
+
+typedef union
+{
+ gvu8 u8;
+ gvu16 u16;
+ gvu32 u32;
+ gvu64 u64;
+ gvs8 s8;
+ gvs16 s16;
+ gvs32 s32;
+ gvs64 s64;
+ // this may be tempting, but it causes gcc to do lots of stack spills
+ //gvhreg h[2];
+} gvreg;
+
+#if defined(__ARM_NEON) || defined(__ARM_NEON__)
+#include <arm_neon.h>
+
+#define gvaddhn_u32(d, a, b) d.u16 = vaddhn_u32(a.u32, b.u32)
+#define gvaddw_s32(d, a, b) d.s64 = vaddw_s32(a.s64, b.s32)
+#define gvabsq_s32(d, s) d.s32 = vabsq_s32(s.s32)
+#define gvbic_n_u16(d, n) d.u16 = vbic_u16(d.u16, vmov_n_u16(n))
+#define gvbifq(d, a, b) d.u8 = vbslq_u8(b.u8, d.u8, a.u8)
+#define gvbit(d, a, b) d.u8 = vbsl_u8(b.u8, a.u8, d.u8)
+#define gvceqq_u16(d, a, b) d.u16 = vceqq_u16(a.u16, b.u16)
+#define gvcgt_s16(d, a, b) d.u16 = vcgt_s16(a.s16, b.s16)
+#define gvclt_s16(d, a, b) d.u16 = vclt_s16(a.s16, b.s16)
+#define gvcreate_s32(d, a, b) d.s32 = vcreate_s32((u32)(a) | ((u64)(b) << 32))
+#define gvcreate_u32(d, a, b) d.u32 = vcreate_u32((u32)(a) | ((u64)(b) << 32))
+#define gvcreate_s64(d, s) d.s64 = (gvhs64)vcreate_s64(s)
+#define gvcreate_u64(d, s) d.u64 = (gvhu64)vcreate_u64(s)
+#define gvcombine_u16(d, l, h) d.u16 = vcombine_u16(l.u16, h.u16)
+#define gvcombine_u32(d, l, h) d.u32 = vcombine_u32(l.u32, h.u32)
+#define gvcombine_s64(d, l, h) d.s64 = vcombine_s64((int64x1_t)l.s64, (int64x1_t)h.s64)
+#define gvdup_l_u8(d, s, l) d.u8 = vdup_lane_u8(s.u8, l)
+#define gvdup_l_u16(d, s, l) d.u16 = vdup_lane_u16(s.u16, l)
+#define gvdup_l_u32(d, s, l) d.u32 = vdup_lane_u32(s.u32, l)
+#define gvdupq_l_s64(d, s, l) d.s64 = vdupq_lane_s64((int64x1_t)s.s64, l)
+#define gvdupq_l_u32(d, s, l) d.u32 = vdupq_lane_u32(s.u32, l)
+#define gvdup_n_s64(d, n) d.s64 = vdup_n_s64(n)
+#define gvdup_n_u8(d, n) d.u8 = vdup_n_u8(n)
+#define gvdup_n_u16(d, n) d.u16 = vdup_n_u16(n)
+#define gvdup_n_u32(d, n) d.u32 = vdup_n_u32(n)
+#define gvdupq_n_u16(d, n) d.u16 = vdupq_n_u16(n)
+#define gvdupq_n_u32(d, n) d.u32 = vdupq_n_u32(n)
+#define gvdupq_n_s64(d, n) d.s64 = vdupq_n_s64(n)
+#define gvhaddq_u16(d, a, b) d.u16 = vhaddq_u16(a.u16, b.u16)
+#define gvmax_s16(d, a, b) d.s16 = vmax_s16(a.s16, b.s16)
+#define gvmin_s16(d, a, b) d.s16 = vmin_s16(a.s16, b.s16)
+#define gvminq_u8(d, a, b) d.u8 = vminq_u8(a.u8, b.u8)
+#define gvminq_u16(d, a, b) d.u16 = vminq_u16(a.u16, b.u16)
+#define gvmla_s32(d, a, b) d.s32 = vmla_s32(d.s32, a.s32, b.s32)
+#define gvmla_u32(d, a, b) d.u32 = vmla_u32(d.u32, a.u32, b.u32)
+#define gvmlaq_s32(d, a, b) d.s32 = vmlaq_s32(d.s32, a.s32, b.s32)
+#define gvmlaq_u32(d, a, b) d.u32 = vmlaq_u32(d.u32, a.u32, b.u32)
+#define gvmlal_s32(d, a, b) d.s64 = vmlal_s32(d.s64, a.s32, b.s32)
+#define gvmlal_u8(d, a, b) d.u16 = vmlal_u8(d.u16, a.u8, b.u8)
+#define gvmlsq_s32(d, a, b) d.s32 = vmlsq_s32(d.s32, a.s32, b.s32)
+#define gvmlsq_l_s32(d, a, b, l) d.s32 = vmlsq_lane_s32(d.s32, a.s32, b.s32, l)
+#define gvmov_l_s32(d, s, l) d.s32 = vset_lane_s32(s, d.s32, l)
+#define gvmov_l_u32(d, s, l) d.u32 = vset_lane_u32(s, d.u32, l)
+#define gvmovl_u8(d, s) d.u16 = vmovl_u8(s.u8)
+#define gvmovl_s32(d, s) d.s64 = vmovl_s32(s.s32)
+#define gvmovn_u16(d, s) d.u8 = vmovn_u16(s.u16)
+#define gvmovn_u32(d, s) d.u16 = vmovn_u32(s.u32)
+#define gvmovn_u64(d, s) d.u32 = vmovn_u64(s.u64)
+#define gvmul_s32(d, a, b) d.s32 = vmul_s32(a.s32, b.s32)
+#define gvmull_s16(d, a, b) d.s32 = vmull_s16(a.s16, b.s16)
+#define gvmull_s32(d, a, b) d.s64 = vmull_s32(a.s32, b.s32)
+#define gvmull_u8(d, a, b) d.u16 = vmull_u8(a.u8, b.u8)
+#define gvmull_l_u32(d, a, b, l) d.u64 = vmull_lane_u32(a.u32, b.u32, l)
+#define gvmlsl_s16(d, a, b) d.s32 = vmlsl_s16(d.s32, a.s16, b.s16)
+#define gvneg_s32(d, s) d.s32 = vneg_s32(s.s32)
+#define gvqadd_u8(d, a, b) d.u8 = vqadd_u8(a.u8, b.u8)
+#define gvqsub_u8(d, a, b) d.u8 = vqsub_u8(a.u8, b.u8)
+#define gvshl_u16(d, a, b) d.u16 = vshl_u16(a.u16, b.s16)
+#define gvshlq_s64(d, a, b) d.s64 = vshlq_s64(a.s64, b.s64)
+#define gvshlq_u32(d, a, b) d.u32 = vshlq_u32(a.u32, b.s32)
+#define gvshlq_u64(d, a, b) d.u64 = vshlq_u64(a.u64, b.s64)
+#define gvshrq_n_s16(d, s, n) d.s16 = vshrq_n_s16(s.s16, n)
+#define gvshrq_n_u16(d, s, n) d.u16 = vshrq_n_u16(s.u16, n)
+#define gvshl_n_u32(d, s, n) d.u32 = vshl_n_u32(s.u32, n)
+#define gvshlq_n_u16(d, s, n) d.u16 = vshlq_n_u16(s.u16, n)
+#define gvshlq_n_u32(d, s, n) d.u32 = vshlq_n_u32(s.u32, n)
+#define gvshll_n_s8(d, s, n) d.s16 = vshll_n_s8(s.s8, n)
+#define gvshll_n_u8(d, s, n) d.u16 = vshll_n_u8(s.u8, n)
+#define gvshll_n_u16(d, s, n) d.u32 = vshll_n_u16(s.u16, n)
+#define gvshr_n_u8(d, s, n) d.u8 = vshr_n_u8(s.u8, n)
+#define gvshr_n_u16(d, s, n) d.u16 = vshr_n_u16(s.u16, n)
+#define gvshr_n_u32(d, s, n) d.u32 = vshr_n_u32(s.u32, n)
+#define gvshr_n_u64(d, s, n) d.u64 = (gvhu64)vshr_n_u64((uint64x1_t)s.u64, n)
+#define gvshrn_n_s64(d, s, n) d.s32 = vshrn_n_s64(s.s64, n)
+#define gvshrn_n_u16(d, s, n) d.u8 = vshrn_n_u16(s.u16, n)
+#define gvshrn_n_u32(d, s, n) d.u16 = vshrn_n_u32(s.u32, n)
+#define gvsli_n_u8(d, s, n) d.u8 = vsli_n_u8(d.u8, s.u8, n)
+#define gvsri_n_u8(d, s, n) d.u8 = vsri_n_u8(d.u8, s.u8, n)
+#define gvtstq_u16(d, a, b) d.u16 = vtstq_u16(a.u16, b.u16)
+#define gvqshrun_n_s16(d, s, n) d.u8 = vqshrun_n_s16(s.s16, n)
+#define gvqsubq_u8(d, a, b) d.u8 = vqsubq_u8(a.u8, b.u8)
+#define gvqsubq_u16(d, a, b) d.u16 = vqsubq_u16(a.u16, b.u16)
+
+#define gvget_lo(d, s) d.u16 = vget_low_u16(s.u16)
+#define gvget_hi(d, s) d.u16 = vget_high_u16(s.u16)
+#define gvlo(s) ({gvhreg t_; gvget_lo(t_, s); t_;})
+#define gvhi(s) ({gvhreg t_; gvget_hi(t_, s); t_;})
+
+#define gvset_lo(d, s) d.u16 = vcombine_u16(s.u16, gvhi(d).u16)
+#define gvset_hi(d, s) d.u16 = vcombine_u16(gvlo(d).u16, s.u16)
+
+#define gvtbl2_u8(d, a, b) { \
+ uint8x8x2_t v_; \
+ v_.val[0] = vget_low_u8(a.u8); v_.val[1] = vget_high_u8(a.u8); \
+ d.u8 = vtbl2_u8(v_, b.u8); \
+}
+
+#define gvzip_u8(d, a, b) { \
+ uint8x8x2_t v_ = vzip_u8(a.u8, b.u8); \
+ d.u8 = vcombine_u8(v_.val[0], v_.val[1]); \
+}
+#define gvzipq_u16(d0, d1, s0, s1) { \
+ uint16x8x2_t v_ = vzipq_u16(s0.u16, s1.u16); \
+ d0.u16 = v_.val[0]; d1.u16 = v_.val[1]; \
+}
+
+#define gvld1_u8(d, s) d.u8 = vld1_u8(s)
+#define gvld1_u32(d, s) d.u32 = vld1_u32((const u32 *)(s))
+#define gvld1q_u8(d, s) d.u8 = vld1q_u8(s)
+#define gvld1q_u16(d, s) d.u16 = vld1q_u16(s)
+#define gvld1q_u32(d, s) d.u32 = vld1q_u32((const u32 *)(s))
+#define gvld2_dup(v0, v1, p) { \
+ uint8x8x2_t v_ = vld2_dup_u8(p); \
+ v0.u8 = v_.val[0]; v1.u8 = v_.val[1]; \
+}
+#define gvld2q_u8(v0, v1, p) { \
+ uint8x16x2_t v_ = vld2q_u8(p); \
+ v0.u8 = v_.val[0]; v1.u8 = v_.val[1]; \
+}
+
+#define gvst1_u8(v, p) \
+ vst1_u8(p, v.u8)
+#define gvst1q_u16(v, p) \
+ vst1q_u16(p, v.u16)
+#define gvst1q_inc_u32(v, p, i) { \
+ vst1q_u32((u32 *)(p), v.u32); \
+ p += (i) / sizeof(*p); \
+}
+#define gvst2_u8(v0, v1, p) { \
+ uint8x8x2_t v_; \
+ v_.val[0] = v0.u8; v_.val[1] = v1.u8; \
+ vst2_u8(p, v_); \
+}
+#define gvst2_u16(v0, v1, p) { \
+ uint16x4x2_t v_; \
+ v_.val[0] = v0.u16; v_.val[1] = v1.u16; \
+ vst2_u16(p, v_); \
+}
+#define gvst2q_u8(v0, v1, p) { \
+ uint8x16x2_t v_; \
+ v_.val[0] = v0.u8; v_.val[1] = v1.u8; \
+ vst2q_u8(p, v_); \
+}
+#define gvst4_4_inc_u32(v0, v1, v2, v3, p, i) { \
+ uint32x2x4_t v_; \
+ v_.val[0] = v0.u32; v_.val[1] = v1.u32; v_.val[2] = v2.u32; v_.val[3] = v3.u32; \
+ vst4_u32(p, v_); p += (i) / sizeof(*p); \
+}
+#define gvst4_pi_u16(v0, v1, v2, v3, p) { \
+ uint16x4x4_t v_; \
+ v_.val[0] = v0.u16; v_.val[1] = v1.u16; v_.val[2] = v2.u16; v_.val[3] = v3.u16; \
+ vst4_u16((u16 *)(p), v_); p += sizeof(v_) / sizeof(*p); \
+}
+#define gvst1q_pi_u32(v, p) \
+ gvst1q_inc_u32(v, p, sizeof(v))
+// could use vst1q_u32_x2 but that's not always available
+#define gvst1q_2_pi_u32(v0, v1, p) { \
+ gvst1q_inc_u32(v0, p, sizeof(v0)); \
+ gvst1q_inc_u32(v1, p, sizeof(v1)); \
+}
+
+/* notes:
+ - gcc > 9: (arm32) int64x1_t type produces ops on gp regs
+ (also u64 __attribute__((vector_size(8))) :( )
+ - gcc <11: (arm32) handles '<vec> == 0' poorly
+*/
+
+/*
+#elif defined(__SSE2__)
+#include <x86intrin.h>
+*/
+#else
+#error "arch not supported or SIMD support was not enabled by your compiler"
+#endif
+
+// the below have intrinsics but they evaluate to basic operations on both gcc and clang
+#define gvadd_s64(d, a, b) d.s64 = a.s64 + b.s64
+#define gvadd_u8(d, a, b) d.u8 = a.u8 + b.u8
+#define gvadd_u16(d, a, b) d.u16 = a.u16 + b.u16
+#define gvadd_u32(d, a, b) d.u32 = a.u32 + b.u32
+#define gvaddq_s64 gvadd_s64
+#define gvaddq_u16 gvadd_u16
+#define gvaddq_u32 gvadd_u32
+#define gvand(d, a, b) d.u32 = a.u32 & b.u32
+#define gvbic(d, a, b) d.u32 = a.u32 & ~b.u32
+#define gvbicq gvbic
+#define gveor(d, a, b) d.u32 = a.u32 ^ b.u32
+#define gveorq gveor
+#define gvceqz_u16(d, s) d.u16 = s.u16 == 0
+#define gvceqzq_u16 gvceqz_u16
+#define gvcltz_s16(d, s) d.s16 = s.s16 < 0
+#define gvcltzq_s16 gvcltz_s16
+#define gvsub_u16(d, a, b) d.u16 = a.u16 - b.u16
+#define gvsub_u32(d, a, b) d.u32 = a.u32 - b.u32
+#define gvsubq_u16 gvsub_u16
+#define gvsubq_u32 gvsub_u32
+#define gvorr(d, a, b) d.u32 = a.u32 | b.u32
+#define gvorrq gvorr
+
+#if defined(__arm__)
+
+#define gssub16(d, a, b) asm("ssub16 %0,%1,%2" : "=r"(d) : "r"(a), "r"(b))
+#define gsmusdx(d, a, b) asm("smusdx %0,%1,%2" : "=r"(d) : "r"(a), "r"(b))
+
+#if 0
+// gcc/config/arm/arm.c
+#undef gvadd_s64
+#define gvadd_s64(d, a, b) asm("vadd.i64 %P0,%P1,%P2" : "=w"(d.s64) : "w"(a.s64), "w"(b.s64))
+#endif
+
+#else
+
+#define gssub16(d, a, b) d = (u16)((a) - (b)) | ((((a) >> 16) - ((b) >> 16)) << 16)
+#define gsmusdx(d, a, b) d = ((s32)(s16)(a) * ((s32)(b) >> 16)) \
+ - (((s32)(a) >> 16) * (s16)(b))
+
+#endif
+
+// for compatibility with the original psx_gpu.c code
+#define vec_2x64s gvreg
+#define vec_2x64u gvreg
+#define vec_4x32s gvreg
+#define vec_4x32u gvreg
+#define vec_8x16s gvreg
+#define vec_8x16u gvreg
+#define vec_16x8s gvreg
+#define vec_16x8u gvreg
+#define vec_1x64s gvhreg
+#define vec_1x64u gvhreg
+#define vec_2x32s gvhreg
+#define vec_2x32u gvhreg
+#define vec_4x16s gvhreg
+#define vec_4x16u gvhreg
+#define vec_8x8s gvhreg
+#define vec_8x8u gvhreg
+
+#if 0
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+static int ccount;
+void cmpp(const char *name, const void *a_, const void *b_, size_t len)
+{
+ const uint32_t *a = a_, *b = b_, masks[] = { 0, 0xff, 0xffff, 0xffffff };
+ size_t i, left;
+ uint32_t mask;
+ for (i = 0; i < (len + 3)/4; i++) {
+ left = len - i*4;
+ mask = left >= 4 ? ~0u : masks[left];
+ if ((a[i] ^ b[i]) & mask) {
+ printf("%s: %08x %08x [%03zx/%zu] #%d\n",
+ name, a[i] & mask, b[i] & mask, i*4, i, ccount);
+ exit(1);
+ }
+ }
+ ccount++;
+}
+#define ccmpf(n) cmpp(#n, &psx_gpu->n, &n##_c, sizeof(n##_c))
+#define ccmpa(n,c) cmpp(#n, &psx_gpu->n, &n##_c, sizeof(n##_c[0]) * c)
+
+void dump_r_(const char *name, void *dump, int is_q)
+{
+ unsigned long long *u = dump;
+ //if (ccount > 1) return;
+ printf("%10s %016llx ", name, u[0]);
+ if (is_q)
+ printf("%016llx", u[1]);
+ puts("");
+}
+void __attribute__((noinline,noclone)) dump_r_d(const char *name, void *dump)
+{ dump_r_(name, dump, 0); }
+void __attribute__((noinline,noclone)) dump_r_q(const char *name, void *dump)
+{ dump_r_(name, dump, 1); }
+#define dumprd(n) { u8 dump_[8]; gvst1_u8(n, dump_); dump_r_d(#n, dump_); }
+#define dumprq(n) { u16 dump_[8]; gvst1q_u16(n, dump_); dump_r_q(#n, dump_); }
+#endif
+
+void compute_all_gradients(psx_gpu_struct * __restrict__ psx_gpu,
+ const vertex_struct * __restrict__ a, const vertex_struct * __restrict__ b,
+ const vertex_struct * __restrict__ c)
+{
+ union { double d; struct { u32 l; u32 h; } i; } divident, divider;
+ union { double d; gvhreg v; } d30;
+
+#if 0
+ compute_all_gradients_(psx_gpu, a, b, c);
+ return;
+#endif
+ // First compute the triangle area reciprocal and shift. The division will
+ // happen concurrently with much of the work which follows.
+
+ // load exponent of 62 into upper half of double
+ u32 shift = __builtin_clz(psx_gpu->triangle_area);
+ u32 triangle_area_normalized = psx_gpu->triangle_area << shift;
+
+ // load area normalized into lower half of double
+ divident.i.l = triangle_area_normalized >> 10;
+ divident.i.h = (62 + 1023) << 20;
+
+ divider.i.l = triangle_area_normalized << 20;
+ divider.i.h = ((1022 + 31) << 20) + (triangle_area_normalized >> 11);
+
+ d30.d = divident.d / divider.d; // d30 = ((1 << 62) + ta_n) / ta_n
+
+ // ((x1 - x0) * (y2 - y1)) - ((x2 - x1) * (y1 - y0)) =
+ // ( d0 * d1 ) - ( d2 * d3 ) =
+ // ( m0 ) - ( m1 ) = gradient
+
+ // This is split to do 12 elements at a time over three sets: a, b, and c.
+ // Technically we only need to do 10 elements (uvrgb_x and uvrgb_y), so
+ // two of the slots are unused.
+
+ // Inputs are all 16-bit signed. The m0/m1 results are 32-bit signed, as
+ // is g.
+
+ // First type is: uvrg bxxx xxxx
+ // Second type is: yyyy ybyy uvrg
+ // Since x_a and y_c are the same the same variable is used for both.
+
+ gvreg v0;
+ gvreg v1;
+ gvreg v2;
+ gvreg uvrg_xxxx0;
+ gvreg uvrg_xxxx1;
+ gvreg uvrg_xxxx2;
+
+ gvreg y0_ab;
+ gvreg y1_ab;
+ gvreg y2_ab;
+
+ gvreg d0_ab;
+ gvreg d1_ab;
+ gvreg d2_ab;
+ gvreg d3_ab;
+
+ gvreg ga_uvrg_x;
+ gvreg ga_uvrg_y;
+ gvreg gw_rg_x;
+ gvreg gw_rg_y;
+ gvreg w_mask;
+ gvreg r_shift;
+ gvreg uvrg_dx2, uvrg_dx3;
+ gvreg uvrgb_phase;
+ gvhreg zero, tmp_lo, tmp_hi;
+
+ gvld1q_u8(v0, (u8 *)a); // v0 = { uvrg0, b0, x0, y0 }
+ gvld1q_u8(v1, (u8 *)b); // v1 = { uvrg1, b1, x1, y1 }
+ gvld1q_u8(v2, (u8 *)c); // v2 = { uvrg2, b2, x2, y2 }
+
+ gvmovl_u8(uvrg_xxxx0, gvlo(v0)); // uvrg_xxxx0 = { uv0, rg0, b0-, -- }
+ gvmovl_u8(uvrg_xxxx1, gvlo(v1)); // uvrg_xxxx1 = { uv1, rg1, b1-, -- }
+ gvmovl_u8(uvrg_xxxx2, gvlo(v2)); // uvrg_xxxx2 = { uv2, rg2, b2-, -- }
+
+ gvdup_l_u16(tmp_lo, gvhi(v0), 1); // yyyy0 = { yy0, yy0 }
+ gvcombine_u16(y0_ab, tmp_lo, gvlo(uvrg_xxxx0));
+
+ gvdup_l_u16(tmp_lo, gvhi(v0), 0); // xxxx0 = { xx0, xx0 }
+ gvset_hi(uvrg_xxxx0, tmp_lo);
+
+ u32 x1_x2 = (u16)b->x | (c->x << 16); // x1_x2 = { x1, x2 }
+ u32 x0_x1 = (u16)a->x | (b->x << 16); // x0_x1 = { x0, x1 }
+
+ gvdup_l_u16(tmp_lo, gvhi(v1), 1); // yyyy1 = { yy1, yy1 }
+ gvcombine_u16(y1_ab, tmp_lo, gvlo(uvrg_xxxx1));
+
+ gvdup_l_u16(tmp_lo, gvhi(v1), 0); // xxxx1 = { xx1, xx1 }
+ gvset_hi(uvrg_xxxx1, tmp_lo);
+
+ gvdup_l_u16(tmp_lo, gvhi(v2), 1); // yyyy2 = { yy2, yy2 }
+ gvcombine_u16(y2_ab, tmp_lo, gvlo(uvrg_xxxx2));
+
+ gvdup_l_u16(tmp_lo, gvhi(v2), 0); // xxxx2 = { xx2, xx2 }
+ gvset_hi(uvrg_xxxx2, tmp_lo);
+
+ u32 y0_y1 = (u16)a->y | (b->y << 16); // y0_y1 = { y0, y1 }
+ u32 y1_y2 = (u16)b->y | (c->y << 16); // y1_y2 = { y1, y2 }
+
+ gvsubq_u16(d0_ab, uvrg_xxxx1, uvrg_xxxx0);
+
+ u32 b1_b2 = b->b | (c->b << 16); // b1_b2 = { b1, b2 }
+
+ gvsubq_u16(d2_ab, uvrg_xxxx2, uvrg_xxxx1);
+
+ gvsubq_u16(d1_ab, y2_ab, y1_ab);
+
+ u32 b0_b1 = a->b | (b->b << 16); // b0_b1 = { b0, b1 }
+
+ u32 dx, dy, db;
+ gssub16(dx, x1_x2, x0_x1); // dx = { x1 - x0, x2 - x1 }
+ gssub16(dy, y1_y2, y0_y1); // dy = { y1 - y0, y2 - y1 }
+ gssub16(db, b1_b2, b0_b1); // db = { b1 - b0, b2 - b1 }
+
+ u32 ga_by, ga_bx;
+ gvsubq_u16(d3_ab, y1_ab, y0_ab);
+ gsmusdx(ga_by, dx, db); // ga_by = ((x1 - x0) * (b2 - b1)) -
+ // ((x2 - X1) * (b1 - b0))
+ gvmull_s16(ga_uvrg_x, gvlo(d0_ab), gvlo(d1_ab));
+ gsmusdx(ga_bx, db, dy); // ga_bx = ((b1 - b0) * (y2 - y1)) -
+ // ((b2 - b1) * (y1 - y0))
+ gvmlsl_s16(ga_uvrg_x, gvlo(d2_ab), gvlo(d3_ab));
+ u32 gs_bx = (s32)ga_bx >> 31; // movs
+
+ gvmull_s16(ga_uvrg_y, gvhi(d0_ab), gvhi(d1_ab));
+ if ((s32)gs_bx < 0) ga_bx = -ga_bx; // rsbmi
+
+ gvmlsl_s16(ga_uvrg_y, gvhi(d2_ab), gvhi(d3_ab));
+ u32 gs_by = (s32)ga_by >> 31; // movs
+
+ gvhreg d0;
+ gvshr_n_u64(d0, d30.v, 22); // note: on "d30 >> 22" gcc generates junk code
+
+ gvdupq_n_u32(uvrgb_phase, psx_gpu->uvrgb_phase);
+ u32 b_base = psx_gpu->uvrgb_phase + (a->b << 16);
+
+ if ((s32)gs_by < 0) ga_by = -ga_by; // rsbmi
+ gvreg gs_uvrg_x, gs_uvrg_y;
+ gs_uvrg_x.s32 = ga_uvrg_x.s32 < 0; // gs_uvrg_x = ga_uvrg_x < 0
+ gs_uvrg_y.s32 = ga_uvrg_y.s32 < 0; // gs_uvrg_y = ga_uvrg_y < 0
+
+ gvdupq_n_u32(w_mask, -psx_gpu->triangle_winding); // w_mask = { -w, -w, -w, -w }
+ shift -= 62 - 12; // shift -= (62 - FIXED_BITS)
+
+ gvreg uvrg_base;
+ gvshll_n_u16(uvrg_base, gvlo(uvrg_xxxx0), 16); // uvrg_base = uvrg0 << 16
+ gvdupq_n_u32(r_shift, shift); // r_shift = { shift, shift, shift, shift }
+
+ gvaddq_u32(uvrg_base, uvrg_base, uvrgb_phase);
+ gvabsq_s32(ga_uvrg_x, ga_uvrg_x); // ga_uvrg_x = abs(ga_uvrg_x)
+
+ u32 area_r_s = d0.u32[0]; // area_r_s = triangle_reciprocal
+ gvabsq_s32(ga_uvrg_y, ga_uvrg_y); // ga_uvrg_y = abs(ga_uvrg_y)
+
+ gvmull_l_u32(gw_rg_x, gvhi(ga_uvrg_x), d0, 0);
+ gvmull_l_u32(ga_uvrg_x, gvlo(ga_uvrg_x), d0, 0);
+ gvmull_l_u32(gw_rg_y, gvhi(ga_uvrg_y), d0, 0);
+ gvmull_l_u32(ga_uvrg_y, gvlo(ga_uvrg_y), d0, 0);
+
+ gvshlq_u64(gw_rg_x, gw_rg_x, r_shift);
+ gvshlq_u64(ga_uvrg_x, ga_uvrg_x, r_shift);
+ gvshlq_u64(gw_rg_y, gw_rg_y, r_shift);
+ gvshlq_u64(ga_uvrg_y, ga_uvrg_y, r_shift);
+
+ gveorq(gs_uvrg_x, gs_uvrg_x, w_mask);
+ gvmovn_u64(tmp_lo, ga_uvrg_x);
+
+ gveorq(gs_uvrg_y, gs_uvrg_y, w_mask);
+ gvmovn_u64(tmp_hi, gw_rg_x);
+
+ gvcombine_u32(ga_uvrg_x, tmp_lo, tmp_hi);
+
+ gveorq(ga_uvrg_x, ga_uvrg_x, gs_uvrg_x);
+ gvmovn_u64(tmp_lo, ga_uvrg_y);
+
+ gvsubq_u32(ga_uvrg_x, ga_uvrg_x, gs_uvrg_x);
+ gvmovn_u64(tmp_hi, gw_rg_y);
+
+ gvcombine_u32(ga_uvrg_y, tmp_lo, tmp_hi);
+
+ gveorq(ga_uvrg_y, ga_uvrg_y, gs_uvrg_y);
+ ga_bx = ga_bx << 13;
+
+ gvsubq_u32(ga_uvrg_y, ga_uvrg_y, gs_uvrg_y);
+ ga_by = ga_by << 13;
+
+ u32 gw_bx_h, gw_by_h;
+ gw_bx_h = (u64)ga_bx * area_r_s >> 32;
+
+ gvshlq_n_u32(ga_uvrg_x, ga_uvrg_x, 4);
+ gvshlq_n_u32(ga_uvrg_y, ga_uvrg_y, 4);
+
+ gw_by_h = (u64)ga_by * area_r_s >> 32;
+ gvdup_n_u32(tmp_lo, a->x);
+ gvmlsq_l_s32(uvrg_base, ga_uvrg_x, tmp_lo, 0);
+
+ gs_bx = gs_bx ^ -psx_gpu->triangle_winding;
+ gvaddq_u32(uvrg_dx2, ga_uvrg_x, ga_uvrg_x);
+
+ gs_by = gs_by ^ -psx_gpu->triangle_winding;
+
+ u32 r11 = -shift; // r11 = negative shift for scalar lsr
+ u32 *store_a = psx_gpu->uvrg.e;
+ r11 = r11 - (32 - 13);
+ u32 *store_b = store_a + 16 / sizeof(u32);
+
+ gvaddq_u32(uvrg_dx3, uvrg_dx2, ga_uvrg_x);
+ gvst1q_inc_u32(uvrg_base, store_a, 32);
+
+ gvst1q_inc_u32(ga_uvrg_x, store_b, 32);
+ u32 g_bx = (u32)gw_bx_h >> r11;
+
+ gvst1q_inc_u32(ga_uvrg_y, store_a, 32);
+ u32 g_by = (u32)gw_by_h >> r11;
+
+ gvdup_n_u32(zero, 0);
+
+ gvst4_4_inc_u32(zero, gvlo(ga_uvrg_x), gvlo(uvrg_dx2), gvlo(uvrg_dx3), store_b, 32);
+ g_bx = g_bx ^ gs_bx;
+
+ gvst4_4_inc_u32(zero, gvhi(ga_uvrg_x), gvhi(uvrg_dx2), gvhi(uvrg_dx3), store_b, 32);
+ g_bx = g_bx - gs_bx;
+
+ g_bx = g_bx << 4;
+ g_by = g_by ^ gs_by;
+
+ b_base -= g_bx * a->x;
+ g_by = g_by - gs_by;
+
+ g_by = g_by << 4;
+
+ u32 g_bx2 = g_bx + g_bx;
+ u32 g_bx3 = g_bx + g_bx2;
+
+ // 112
+ store_b[0] = 0;
+ store_b[1] = g_bx;
+ store_b[2] = g_bx2;
+ store_b[3] = g_bx3;
+ store_b[4] = b_base;
+ store_b[5] = g_by; // 132
+}
+
+#define setup_spans_debug_check(span_edge_data_element) \
+
+#define setup_spans_prologue_alternate_yes() \
+ vec_2x64s alternate_x; \
+ vec_2x64s alternate_dx_dy; \
+ vec_4x32s alternate_x_32; \
+ vec_2x32s alternate_x_16; \
+ \
+ vec_4x16u alternate_select; \
+ vec_4x16s y_mid_point; \
+ \
+ s32 y_b = v_b->y; \
+ s64 edge_alt; \
+ s32 edge_dx_dy_alt; \
+ u32 edge_shift_alt \
+
+#define setup_spans_prologue_alternate_no() \
+
+#define setup_spans_prologue(alternate_active) \
+ edge_data_struct *span_edge_data; \
+ vec_4x32u *span_uvrg_offset; \
+ u32 *span_b_offset; \
+ \
+ s32 clip; \
+ vec_4x32u v_clip; \
+ \
+ union { vec_2x64s full; vec_1x64s h[2]; } edges_xy; \
+ vec_2x32s edges_dx_dy; \
+ vec_2x32u edge_shifts; \
+ \
+ vec_2x64s left_x, right_x; \
+ vec_2x64s left_dx_dy, right_dx_dy; \
+ vec_4x32s left_x_32, right_x_32; \
+ vec_2x32s left_x_32_lo, right_x_32_lo; \
+ vec_2x32s left_x_32_hi, right_x_32_hi; \
+ vec_4x16s left_right_x_16_lo, left_right_x_16_hi; \
+ vec_4x16s y_x4; \
+ vec_8x16s left_edge; \
+ vec_8x16s right_edge; \
+ vec_4x16u span_shift; \
+ \
+ vec_2x32u c_0x01; \
+ vec_4x16u c_0x04; \
+ vec_4x16u c_0xFFFE; \
+ vec_4x16u c_0x07; \
+ \
+ vec_2x32s x_starts; \
+ vec_2x32s x_ends; \
+ \
+ s32 x_a = v_a->x; \
+ s32 x_b = v_b->x; \
+ s32 x_c = v_c->x; \
+ s32 y_a = v_a->y; \
+ s32 y_c = v_c->y; \
+ \
+ vec_4x32u uvrg; \
+ vec_4x32u uvrg_dy; \
+ u32 b = psx_gpu->b; \
+ u32 b_dy = psx_gpu->b_dy; \
+ const u32 *reciprocal_table = psx_gpu->reciprocal_table_ptr; \
+ \
+ gvld1q_u32(uvrg, psx_gpu->uvrg.e); \
+ gvld1q_u32(uvrg_dy, psx_gpu->uvrg_dy.e); \
+ gvdup_n_u32(c_0x01, 0x01); \
+ setup_spans_prologue_alternate_##alternate_active() \
+
+#define setup_spans_prologue_b() \
+ span_edge_data = psx_gpu->span_edge_data; \
+ span_uvrg_offset = (vec_4x32u *)psx_gpu->span_uvrg_offset; \
+ span_b_offset = psx_gpu->span_b_offset; \
+ \
+ vec_8x16u c_0x0001; \
+ \
+ gvdupq_n_u16(c_0x0001, 0x0001); \
+ gvdupq_n_u16(left_edge, psx_gpu->viewport_start_x); \
+ gvdupq_n_u16(right_edge, psx_gpu->viewport_end_x); \
+ gvaddq_u16(right_edge, right_edge, c_0x0001); \
+ gvdup_n_u16(c_0x04, 0x04); \
+ gvdup_n_u16(c_0x07, 0x07); \
+ gvdup_n_u16(c_0xFFFE, 0xFFFE); \
+
+
+#define compute_edge_delta_x2() \
+{ \
+ vec_2x32s heights; \
+ vec_2x32s height_reciprocals; \
+ vec_2x32s heights_b; \
+ vec_2x32u widths; \
+ \
+ u32 edge_shift = reciprocal_table[height]; \
+ \
+ gvdup_n_u32(heights, height); \
+ gvsub_u32(widths, x_ends, x_starts); \
+ \
+ gvdup_n_u32(edge_shifts, edge_shift); \
+ gvsub_u32(heights_b, heights, c_0x01); \
+ gvshr_n_u32(height_reciprocals, edge_shifts, 10); \
+ \
+ gvmla_s32(heights_b, x_starts, heights); \
+ gvbic_n_u16(edge_shifts, 0xE0); \
+ gvmul_s32(edges_dx_dy, widths, height_reciprocals); \
+ gvmull_s32(edges_xy.full, heights_b, height_reciprocals); \
+} \
+
+#define compute_edge_delta_x3(start_c, height_a, height_b) \
+{ \
+ vec_2x32s heights; \
+ vec_2x32s height_reciprocals; \
+ vec_2x32s heights_b; \
+ vec_2x32u widths; \
+ \
+ u32 width_alt; \
+ s32 height_b_alt; \
+ u32 height_reciprocal_alt; \
+ \
+ gvcreate_u32(heights, height_a, height_b); \
+ gvcreate_u32(edge_shifts, reciprocal_table[height_a], reciprocal_table[height_b]); \
+ \
+ edge_shift_alt = reciprocal_table[height_minor_b]; \
+ \
+ gvsub_u32(widths, x_ends, x_starts); \
+ width_alt = x_c - start_c; \
+ \
+ gvshr_n_u32(height_reciprocals, edge_shifts, 10); \
+ height_reciprocal_alt = edge_shift_alt >> 10; \
+ \
+ gvbic_n_u16(edge_shifts, 0xE0); \
+ edge_shift_alt &= 0x1F; \
+ \
+ gvsub_u32(heights_b, heights, c_0x01); \
+ height_b_alt = height_minor_b - 1; \
+ \
+ gvmla_s32(heights_b, x_starts, heights); \
+ height_b_alt += height_minor_b * start_c; \
+ \
+ gvmull_s32(edges_xy.full, heights_b, height_reciprocals); \
+ edge_alt = (s64)height_b_alt * height_reciprocal_alt; \
+ \
+ gvmul_s32(edges_dx_dy, widths, height_reciprocals); \
+ edge_dx_dy_alt = width_alt * height_reciprocal_alt; \
+} \
+
+
+#define setup_spans_adjust_y_up() \
+ gvsub_u32(y_x4, y_x4, c_0x04) \
+
+#define setup_spans_adjust_y_down() \
+ gvadd_u32(y_x4, y_x4, c_0x04) \
+
+#define setup_spans_adjust_interpolants_up() \
+ gvsubq_u32(uvrg, uvrg, uvrg_dy); \
+ b -= b_dy \
+
+#define setup_spans_adjust_interpolants_down() \
+ gvaddq_u32(uvrg, uvrg, uvrg_dy); \
+ b += b_dy \
+
+
+#define setup_spans_clip_interpolants_increment() \
+ gvmlaq_s32(uvrg, uvrg_dy, v_clip); \
+ b += b_dy * clip \
+
+#define setup_spans_clip_interpolants_decrement() \
+ gvmlsq_s32(uvrg, uvrg_dy, v_clip); \
+ b -= b_dy * clip \
+
+#define setup_spans_clip_alternate_yes() \
+ edge_alt += edge_dx_dy_alt * (s64)(clip) \
+
+#define setup_spans_clip_alternate_no() \
+
+#define setup_spans_clip(direction, alternate_active) \
+{ \
+ gvdupq_n_u32(v_clip, clip); \
+ gvmlal_s32(edges_xy.full, edges_dx_dy, gvlo(v_clip)); \
+ setup_spans_clip_alternate_##alternate_active(); \
+ setup_spans_clip_interpolants_##direction(); \
+} \
+
+
+#define setup_spans_adjust_edges_alternate_no(left_index, right_index) \
+{ \
+ vec_2x64s edge_shifts_64; \
+ union { vec_2x64s full; vec_1x64s h[2]; } edges_dx_dy_64; \
+ vec_1x64s left_x_hi, right_x_hi; \
+ \
+ gvmovl_s32(edge_shifts_64, edge_shifts); \
+ gvshlq_s64(edges_xy.full, edges_xy.full, edge_shifts_64); \
+ \
+ gvmovl_s32(edges_dx_dy_64.full, edges_dx_dy); \
+ gvshlq_s64(edges_dx_dy_64.full, edges_dx_dy_64.full, edge_shifts_64); \
+ \
+ gvdupq_l_s64(left_x, edges_xy.h[left_index], 0); \
+ gvdupq_l_s64(right_x, edges_xy.h[right_index], 0); \
+ \
+ gvdupq_l_s64(left_dx_dy, edges_dx_dy_64.h[left_index], 0); \
+ gvdupq_l_s64(right_dx_dy, edges_dx_dy_64.h[right_index], 0); \
+ \
+ gvadd_s64(left_x_hi, gvlo(left_x), gvlo(left_dx_dy)); \
+ gvadd_s64(right_x_hi, gvlo(right_x), gvlo(right_dx_dy)); \
+ \
+ gvset_hi(left_x, left_x_hi); \
+ gvset_hi(right_x, right_x_hi); \
+ \
+ gvaddq_s64(left_dx_dy, left_dx_dy, left_dx_dy); \
+ gvaddq_s64(right_dx_dy, right_dx_dy, right_dx_dy); \
+} \
+
+#define setup_spans_adjust_edges_alternate_yes(left_index, right_index) \
+{ \
+ setup_spans_adjust_edges_alternate_no(left_index, right_index); \
+ s64 edge_dx_dy_alt_64; \
+ vec_1x64s alternate_x_hi; \
+ \
+ gvdup_n_u16(y_mid_point, y_b); \
+ \
+ edge_alt <<= edge_shift_alt; \
+ edge_dx_dy_alt_64 = (s64)edge_dx_dy_alt << edge_shift_alt; \
+ \
+ gvdupq_n_s64(alternate_x, edge_alt); \
+ gvdupq_n_s64(alternate_dx_dy, edge_dx_dy_alt_64); \
+ \
+ gvadd_s64(alternate_x_hi, gvlo(alternate_x), gvlo(alternate_dx_dy)); \
+ gvaddq_s64(alternate_dx_dy, alternate_dx_dy, alternate_dx_dy); \
+ gvset_hi(alternate_x, alternate_x_hi); \
+} \
+
+
+#define setup_spans_y_select_up() \
+ gvclt_s16(alternate_select, y_x4, y_mid_point) \
+
+#define setup_spans_y_select_down() \
+ gvcgt_s16(alternate_select, y_x4, y_mid_point) \
+
+#define setup_spans_y_select_alternate_yes(direction) \
+ setup_spans_y_select_##direction() \
+
+#define setup_spans_y_select_alternate_no(direction) \
+
+#define setup_spans_alternate_select_left() \
+ gvbit(left_right_x_16_lo, alternate_x_16, alternate_select); \
+
+#define setup_spans_alternate_select_right() \
+ gvbit(left_right_x_16_hi, alternate_x_16, alternate_select); \
+
+#define setup_spans_alternate_select_none() \
+
+#define setup_spans_increment_alternate_yes() \
+{ \
+ vec_2x32s alternate_x_32_lo, alternate_x_32_hi; \
+ gvshrn_n_s64(alternate_x_32_lo, alternate_x, 32); \
+ gvaddq_s64(alternate_x, alternate_x, alternate_dx_dy); \
+ gvshrn_n_s64(alternate_x_32_hi, alternate_x, 32); \
+ gvaddq_s64(alternate_x, alternate_x, alternate_dx_dy); \
+ gvcombine_u32(alternate_x_32, alternate_x_32_lo, alternate_x_32_hi); \
+ gvmovn_u32(alternate_x_16, alternate_x_32); \
+} \
+
+#define setup_spans_increment_alternate_no() \
+
+#define setup_spans_set_x4(alternate, direction, alternate_active) \
+{ \
+ gvst1q_pi_u32(uvrg, span_uvrg_offset); \
+ *span_b_offset++ = b; \
+ setup_spans_adjust_interpolants_##direction(); \
+ \
+ gvst1q_pi_u32(uvrg, span_uvrg_offset); \
+ *span_b_offset++ = b; \
+ setup_spans_adjust_interpolants_##direction(); \
+ \
+ gvst1q_pi_u32(uvrg, span_uvrg_offset); \
+ *span_b_offset++ = b; \
+ setup_spans_adjust_interpolants_##direction(); \
+ \
+ gvst1q_pi_u32(uvrg, span_uvrg_offset); \
+ *span_b_offset++ = b; \
+ setup_spans_adjust_interpolants_##direction(); \
+ \
+ gvshrn_n_s64(left_x_32_lo, left_x, 32); \
+ gvshrn_n_s64(right_x_32_lo, right_x, 32); \
+ \
+ gvaddq_s64(left_x, left_x, left_dx_dy); \
+ gvaddq_s64(right_x, right_x, right_dx_dy); \
+ \
+ gvshrn_n_s64(left_x_32_hi, left_x, 32); \
+ gvshrn_n_s64(right_x_32_hi, right_x, 32); \
+ \
+ gvaddq_s64(left_x, left_x, left_dx_dy); \
+ gvaddq_s64(right_x, right_x, right_dx_dy); \
+ \
+ gvcombine_s64(left_x_32, left_x_32_lo, left_x_32_hi); \
+ gvcombine_s64(right_x_32, right_x_32_lo, right_x_32_hi); \
+ \
+ gvmovn_u32(left_right_x_16_lo, left_x_32); \
+ gvmovn_u32(left_right_x_16_hi, right_x_32); \
+ \
+ setup_spans_increment_alternate_##alternate_active(); \
+ setup_spans_y_select_alternate_##alternate_active(direction); \
+ setup_spans_alternate_select_##alternate(); \
+ \
+ gvmax_s16(left_right_x_16_lo, left_right_x_16_lo, gvlo(left_edge)); \
+ gvmax_s16(left_right_x_16_hi, left_right_x_16_hi, gvhi(left_edge)); \
+ gvmin_s16(left_right_x_16_lo, left_right_x_16_lo, gvlo(right_edge)); \
+ gvmin_s16(left_right_x_16_hi, left_right_x_16_hi, gvhi(right_edge)); \
+ \
+ gvsub_u16(left_right_x_16_hi, left_right_x_16_hi, left_right_x_16_lo); \
+ gvadd_u16(left_right_x_16_hi, left_right_x_16_hi, c_0x07); \
+ gvand(span_shift, left_right_x_16_hi, c_0x07); \
+ gvshl_u16(span_shift, c_0xFFFE, span_shift); \
+ gvshr_n_u16(left_right_x_16_hi, left_right_x_16_hi, 3); \
+ \
+ gvst4_pi_u16(left_right_x_16_lo, left_right_x_16_hi, span_shift, y_x4, \
+ span_edge_data); \
+ \
+ setup_spans_adjust_y_##direction(); \
+} \
+
+
+#define setup_spans_alternate_adjust_yes() \
+ edge_alt -= edge_dx_dy_alt * (s64)height_minor_a \
+
+#define setup_spans_alternate_adjust_no() \
+
+
+#define setup_spans_down(left_index, right_index, alternate, alternate_active) \
+ setup_spans_alternate_adjust_##alternate_active(); \
+ if(y_c > psx_gpu->viewport_end_y) \
+ height -= y_c - psx_gpu->viewport_end_y - 1; \
+ \
+ clip = psx_gpu->viewport_start_y - y_a; \
+ if(clip > 0) \
+ { \
+ height -= clip; \
+ y_a += clip; \
+ setup_spans_clip(increment, alternate_active); \
+ } \
+ \
+ setup_spans_prologue_b(); \
+ \
+ if(height > 0) \
+ { \
+ u64 y_x4_ = ((u64)(y_a + 3) << 48) | ((u64)(u16)(y_a + 2) << 32) \
+ | (u32)((y_a + 1) << 16) | (u16)y_a; \
+ gvcreate_u64(y_x4, y_x4_); \
+ setup_spans_adjust_edges_alternate_##alternate_active(left_index, \
+ right_index); \
+ \
+ psx_gpu->num_spans = height; \
+ do \
+ { \
+ setup_spans_set_x4(alternate, down, alternate_active); \
+ height -= 4; \
+ } while(height > 0); \
+ } \
+
+
+#define setup_spans_alternate_pre_increment_yes() \
+ edge_alt += edge_dx_dy_alt \
+
+#define setup_spans_alternate_pre_increment_no() \
+
+#define setup_spans_up_decrement_height_yes() \
+ height-- \
+
+#define setup_spans_up_decrement_height_no() \
+ {} \
+
+#define setup_spans_up(left_index, right_index, alternate, alternate_active) \
+ setup_spans_alternate_adjust_##alternate_active(); \
+ y_a--; \
+ \
+ if(y_c < psx_gpu->viewport_start_y) \
+ height -= psx_gpu->viewport_start_y - y_c; \
+ else \
+ setup_spans_up_decrement_height_##alternate_active(); \
+ \
+ clip = y_a - psx_gpu->viewport_end_y; \
+ if(clip > 0) \
+ { \
+ height -= clip; \
+ y_a -= clip; \
+ setup_spans_clip(decrement, alternate_active); \
+ } \
+ \
+ setup_spans_prologue_b(); \
+ \
+ if(height > 0) \
+ { \
+ u64 y_x4_ = ((u64)(y_a - 3) << 48) | ((u64)(u16)(y_a - 2) << 32) \
+ | (u32)((y_a - 1) << 16) | (u16)y_a; \
+ gvcreate_u64(y_x4, y_x4_); \
+ gvaddw_s32(edges_xy.full, edges_xy.full, edges_dx_dy); \
+ setup_spans_alternate_pre_increment_##alternate_active(); \
+ setup_spans_adjust_edges_alternate_##alternate_active(left_index, \
+ right_index); \
+ setup_spans_adjust_interpolants_up(); \
+ \
+ psx_gpu->num_spans = height; \
+ while(height > 0) \
+ { \
+ setup_spans_set_x4(alternate, up, alternate_active); \
+ height -= 4; \
+ } \
+ } \
+
+#define index_left 0
+#define index_right 1
+
+#define setup_spans_up_up(minor, major) \
+ setup_spans_prologue(yes); \
+ s32 height_minor_a = y_a - y_b; \
+ s32 height_minor_b = y_b - y_c; \
+ s32 height = y_a - y_c; \
+ \
+ gvdup_n_u32(x_starts, x_a); \
+ gvcreate_u32(x_ends, x_c, x_b); \
+ \
+ compute_edge_delta_x3(x_b, height, height_minor_a); \
+ setup_spans_up(index_##major, index_##minor, minor, yes) \
+
+void setup_spans_up_left(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_up_left_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_up_up(left, right)
+}
+
+void setup_spans_up_right(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_up_right_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_up_up(right, left)
+}
+
+#define setup_spans_down_down(minor, major) \
+ setup_spans_prologue(yes); \
+ s32 height_minor_a = y_b - y_a; \
+ s32 height_minor_b = y_c - y_b; \
+ s32 height = y_c - y_a; \
+ \
+ gvdup_n_u32(x_starts, x_a); \
+ gvcreate_u32(x_ends, x_c, x_b); \
+ \
+ compute_edge_delta_x3(x_b, height, height_minor_a); \
+ setup_spans_down(index_##major, index_##minor, minor, yes) \
+
+void setup_spans_down_left(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_down_left_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_down_down(left, right)
+}
+
+void setup_spans_down_right(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_down_right_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_down_down(right, left)
+}
+
+#define setup_spans_up_flat() \
+ s32 height = y_a - y_c; \
+ \
+ compute_edge_delta_x2(); \
+ setup_spans_up(index_left, index_right, none, no) \
+
+void setup_spans_up_a(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_up_a_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_prologue(no);
+
+ gvcreate_u32(x_starts, x_a, x_b);
+ gvdup_n_u32(x_ends, x_c);
+
+ setup_spans_up_flat()
+}
+
+void setup_spans_up_b(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_up_b_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_prologue(no);
+
+ gvdup_n_u32(x_starts, x_a);
+ gvcreate_u32(x_ends, x_b, x_c);
+
+ setup_spans_up_flat()
+}
+
+#define setup_spans_down_flat() \
+ s32 height = y_c - y_a; \
+ \
+ compute_edge_delta_x2(); \
+ setup_spans_down(index_left, index_right, none, no) \
+
+void setup_spans_down_a(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_down_a_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_prologue(no);
+
+ gvcreate_u32(x_starts, x_a, x_b);
+ gvdup_n_u32(x_ends, x_c);
+
+ setup_spans_down_flat()
+}
+
+void setup_spans_down_b(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_down_b_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_prologue(no)
+
+ gvdup_n_u32(x_starts, x_a);
+ gvcreate_u32(x_ends, x_b, x_c);
+
+ setup_spans_down_flat()
+}
+
+void setup_spans_up_down(psx_gpu_struct *psx_gpu, vertex_struct *v_a,
+ vertex_struct *v_b, vertex_struct *v_c)
+{
+#if 0
+ setup_spans_up_down_(psx_gpu, v_a, v_b, v_c);
+ return;
+#endif
+ setup_spans_prologue(no);
+
+ s32 y_b = v_b->y;
+ s64 edge_alt;
+ s32 edge_dx_dy_alt;
+ u32 edge_shift_alt;
+
+ s32 middle_y = y_a;
+ s32 height_minor_a = y_a - y_b;
+ s32 height_minor_b = y_c - y_a;
+ s32 height_major = y_c - y_b;
+
+ vec_2x64s edges_xy_b;
+ vec_1x64s edges_xy_b_left;
+ vec_2x32s edges_dx_dy_b;
+ vec_2x32u edge_shifts_b;
+
+ vec_2x32s height_increment;
+
+ gvcreate_u32(x_starts, x_a, x_c);
+ gvdup_n_u32(x_ends, x_b);
+
+ compute_edge_delta_x3(x_a, height_minor_a, height_major);
+
+ gvcreate_s32(height_increment, 0, height_minor_b);
+
+ gvmlal_s32(edges_xy.full, edges_dx_dy, height_increment);
+
+ gvcreate_s64(edges_xy_b_left, edge_alt);
+ gvcombine_s64(edges_xy_b, edges_xy_b_left, gvhi(edges_xy.full));
+
+ edge_shifts_b = edge_shifts;
+ gvmov_l_u32(edge_shifts_b, edge_shift_alt, 0);
+
+ gvneg_s32(edges_dx_dy_b, edges_dx_dy);
+ gvmov_l_s32(edges_dx_dy_b, edge_dx_dy_alt, 0);
+
+ y_a--;
+
+ if(y_b < psx_gpu->viewport_start_y)
+ height_minor_a -= psx_gpu->viewport_start_y - y_b;
+
+ clip = y_a - psx_gpu->viewport_end_y;
+ if(clip > 0)
+ {
+ height_minor_a -= clip;
+ y_a -= clip;
+ setup_spans_clip(decrement, no);
+ }
+
+ setup_spans_prologue_b();
+
+ if(height_minor_a > 0)
+ {
+ u64 y_x4_ = ((u64)(y_a - 3) << 48) | ((u64)(u16)(y_a - 2) << 32)
+ | (u32)((y_a - 1) << 16) | (u16)y_a;
+ gvcreate_u64(y_x4, y_x4_);
+ gvaddw_s32(edges_xy.full, edges_xy.full, edges_dx_dy);
+ setup_spans_adjust_edges_alternate_no(index_left, index_right);
+ setup_spans_adjust_interpolants_up();
+
+ psx_gpu->num_spans = height_minor_a;
+ while(height_minor_a > 0)
+ {
+ setup_spans_set_x4(none, up, no);
+ height_minor_a -= 4;
+ }
+
+ span_edge_data += height_minor_a;
+ span_uvrg_offset += height_minor_a;
+ span_b_offset += height_minor_a;
+ }
+
+ edges_xy.full = edges_xy_b;
+ edges_dx_dy = edges_dx_dy_b;
+ edge_shifts = edge_shifts_b;
+
+ gvld1q_u32(uvrg, psx_gpu->uvrg.e);
+ b = psx_gpu->b;
+
+ y_a = middle_y;
+
+ if(y_c > psx_gpu->viewport_end_y)
+ height_minor_b -= y_c - psx_gpu->viewport_end_y - 1;
+
+ clip = psx_gpu->viewport_start_y - y_a;
+ if(clip > 0)
+ {
+ height_minor_b -= clip;
+ y_a += clip;
+ setup_spans_clip(increment, no);
+ }
+
+ if(height_minor_b > 0)
+ {
+ u64 y_x4_ = ((u64)(y_a + 3) << 48) | ((u64)(u16)(y_a + 2) << 32)
+ | (u32)((y_a + 1) << 16) | (u16)y_a;
+ gvcreate_u64(y_x4, y_x4_);
+ setup_spans_adjust_edges_alternate_no(index_left, index_right);
+
+ // FIXME: overflow corner case
+ if(psx_gpu->num_spans + height_minor_b == MAX_SPANS)
+ height_minor_b &= ~3;
+
+ psx_gpu->num_spans += height_minor_b;
+ while(height_minor_b > 0)
+ {
+ setup_spans_set_x4(none, down, no);
+ height_minor_b -= 4;
+ }
+ }
+}
+
+
+#define dither_table_entry_normal(value) \
+ (value) \
+
+#define setup_blocks_load_msb_mask_indirect() \
+
+#define setup_blocks_load_msb_mask_direct() \
+ vec_8x16u msb_mask; \
+ gvdupq_n_u16(msb_mask, psx_gpu->mask_msb); \
+
+#define setup_blocks_variables_shaded_textured(target) \
+ vec_4x32u u_block; \
+ vec_4x32u v_block; \
+ vec_4x32u r_block; \
+ vec_4x32u g_block; \
+ vec_4x32u b_block; \
+ vec_4x32u uvrg_dx; \
+ vec_4x32u uvrg_dx4; \
+ vec_4x32u uvrg_dx8; \
+ vec_4x32u uvrg; \
+ vec_16x8u texture_mask; \
+ vec_8x8u texture_mask_lo, texture_mask_hi; \
+ u32 b_dx = psx_gpu->b_block_span.e[1]; \
+ u32 b_dx4 = b_dx << 2; \
+ u32 b_dx8 = b_dx << 3; \
+ u32 b; \
+ \
+ gvld1q_u32(uvrg_dx, psx_gpu->uvrg_dx.e); \
+ gvshlq_n_u32(uvrg_dx4, uvrg_dx, 2); \
+ gvshlq_n_u32(uvrg_dx8, uvrg_dx, 3); \
+ gvld2_dup(texture_mask_lo, texture_mask_hi, &psx_gpu->texture_mask_width); \
+ gvcombine_u16(texture_mask, texture_mask_lo, texture_mask_hi) \
+
+#define setup_blocks_variables_shaded_untextured(target) \
+ vec_4x32u r_block; \
+ vec_4x32u g_block; \
+ vec_4x32u b_block; \
+ vec_4x32u rgb_dx; \
+ vec_2x32u rgb_dx_lo, rgb_dx_hi; \
+ vec_4x32u rgb_dx4; \
+ vec_4x32u rgb_dx8; \
+ vec_4x32u rgb; \
+ vec_2x32u rgb_lo, rgb_hi; \
+ \
+ vec_8x8u d64_0x07; \
+ vec_8x8u d64_1; \
+ vec_8x8u d64_4; \
+ vec_8x8u d64_128; \
+ \
+ gvdup_n_u8(d64_0x07, 0x07); \
+ gvdup_n_u8(d64_1, 1); \
+ gvdup_n_u8(d64_4, 4); \
+ gvdup_n_u8(d64_128, 128); \
+ \
+ gvld1_u32(rgb_dx_lo, &psx_gpu->uvrg_dx.e[2]); \
+ gvcreate_u32(rgb_dx_hi, psx_gpu->b_block_span.e[1], 0); \
+ gvcombine_u32(rgb_dx, rgb_dx_lo, rgb_dx_hi); \
+ gvshlq_n_u32(rgb_dx4, rgb_dx, 2); \
+ gvshlq_n_u32(rgb_dx8, rgb_dx, 3) \
+
+#define setup_blocks_variables_unshaded_textured(target) \
+ vec_4x32u u_block; \
+ vec_4x32u v_block; \
+ vec_2x32u uv_dx; \
+ vec_2x32u uv_dx4; \
+ vec_2x32u uv_dx8; \
+ vec_2x32u uv; \
+ vec_16x8u texture_mask; \
+ vec_8x8u texture_mask_lo, texture_mask_hi; \
+ \
+ gvld1_u32(uv_dx, psx_gpu->uvrg_dx.e); \
+ gvld1_u32(uv, psx_gpu->uvrg.e); \
+ gvshl_n_u32(uv_dx4, uv_dx, 2); \
+ gvshl_n_u32(uv_dx8, uv_dx, 3); \
+ gvld2_dup(texture_mask_lo, texture_mask_hi, &psx_gpu->texture_mask_width); \
+ gvcombine_u16(texture_mask, texture_mask_lo, texture_mask_hi) \
+
+#define setup_blocks_variables_unshaded_untextured_direct() \
+ gvorrq(colors, colors, msb_mask) \
+
+#define setup_blocks_variables_unshaded_untextured_indirect() \
+
+#define setup_blocks_variables_unshaded_untextured(target) \
+ u32 color = psx_gpu->triangle_color; \
+ vec_8x16u colors; \
+ \
+ u32 color_r = color & 0xFF; \
+ u32 color_g = (color >> 8) & 0xFF; \
+ u32 color_b = (color >> 16) & 0xFF; \
+ \
+ color = (color_r >> 3) | ((color_g >> 3) << 5) | \
+ ((color_b >> 3) << 10); \
+ gvdupq_n_u16(colors, color); \
+ setup_blocks_variables_unshaded_untextured_##target() \
+
+#define setup_blocks_span_initialize_dithered_textured() \
+ vec_8x16u dither_offsets; \
+ gvshll_n_s8(dither_offsets, dither_offsets_short, 4) \
+
+#define setup_blocks_span_initialize_dithered_untextured() \
+ vec_8x8u dither_offsets; \
+ gvadd_u8(dither_offsets, dither_offsets_short, d64_4) \
+
+#define setup_blocks_span_initialize_dithered(texturing) \
+ u32 dither_row = psx_gpu->dither_table[y & 0x3]; \
+ u32 dither_shift = (span_edge_data->left_x & 0x3) * 8; \
+ vec_8x8s dither_offsets_short; \
+ \
+ dither_row = \
+ (dither_row >> dither_shift) | (dither_row << (32 - dither_shift)); \
+ gvdup_n_u32(dither_offsets_short, dither_row); \
+ setup_blocks_span_initialize_dithered_##texturing() \
+
+#define setup_blocks_span_initialize_undithered(texturing) \
+
+#define setup_blocks_span_initialize_shaded_textured() \
+{ \
+ u32 left_x = span_edge_data->left_x; \
+ vec_4x32u block_span; \
+ vec_4x32u v_left_x; \
+ \
+ gvld1q_u32(uvrg, span_uvrg_offset); \
+ gvdupq_n_u32(v_left_x, left_x); \
+ gvmlaq_u32(uvrg, uvrg_dx, v_left_x); \
+ b = *span_b_offset; \
+ b += b_dx * left_x; \
+ \
+ gvdupq_l_u32(u_block, gvlo(uvrg), 0); \
+ gvdupq_l_u32(v_block, gvlo(uvrg), 1); \
+ gvdupq_l_u32(r_block, gvhi(uvrg), 0); \
+ gvdupq_l_u32(g_block, gvhi(uvrg), 1); \
+ gvdupq_n_u32(b_block, b); \
+ \
+ gvld1q_u32(block_span, psx_gpu->u_block_span.e); \
+ gvaddq_u32(u_block, u_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->v_block_span.e); \
+ gvaddq_u32(v_block, v_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->r_block_span.e); \
+ gvaddq_u32(r_block, r_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->g_block_span.e); \
+ gvaddq_u32(g_block, g_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->b_block_span.e); \
+ gvaddq_u32(b_block, b_block, block_span); \
+}
+
+#define setup_blocks_span_initialize_shaded_untextured() \
+{ \
+ u32 left_x = span_edge_data->left_x; \
+ u32 *span_uvrg_offset_high = (u32 *)span_uvrg_offset + 2; \
+ vec_4x32u block_span; \
+ vec_4x32u v_left_x; \
+ \
+ gvld1_u32(rgb_lo, span_uvrg_offset_high); \
+ gvcreate_u32(rgb_hi, *span_b_offset, 0); \
+ gvcombine_u32(rgb, rgb_lo, rgb_hi); \
+ gvdupq_n_u32(v_left_x, left_x); \
+ gvmlaq_u32(rgb, rgb_dx, v_left_x); \
+ \
+ gvdupq_l_u32(r_block, gvlo(rgb), 0); \
+ gvdupq_l_u32(g_block, gvlo(rgb), 1); \
+ gvdupq_l_u32(b_block, gvhi(rgb), 0); \
+ \
+ gvld1q_u32(block_span, psx_gpu->r_block_span.e); \
+ gvaddq_u32(r_block, r_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->g_block_span.e); \
+ gvaddq_u32(g_block, g_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->b_block_span.e); \
+ gvaddq_u32(b_block, b_block, block_span); \
+} \
+
+#define setup_blocks_span_initialize_unshaded_textured() \
+{ \
+ u32 left_x = span_edge_data->left_x; \
+ vec_4x32u block_span; \
+ vec_2x32u v_left_x; \
+ \
+ gvld1_u32(uv, span_uvrg_offset); \
+ gvdup_n_u32(v_left_x, left_x); \
+ gvmla_u32(uv, uv_dx, v_left_x); \
+ \
+ gvdupq_l_u32(u_block, uv, 0); \
+ gvdupq_l_u32(v_block, uv, 1); \
+ \
+ gvld1q_u32(block_span, psx_gpu->u_block_span.e); \
+ gvaddq_u32(u_block, u_block, block_span); \
+ gvld1q_u32(block_span, psx_gpu->v_block_span.e); \
+ gvaddq_u32(v_block, v_block, block_span); \
+} \
+
+#define setup_blocks_span_initialize_unshaded_untextured() \
+
+#define setup_blocks_texture_swizzled() \
+{ \
+ vec_8x8u u_saved = u; \
+ gvsli_n_u8(u, v, 4); \
+ gvsri_n_u8(v, u_saved, 4); \
+} \
+
+#define setup_blocks_texture_unswizzled() \
+
+#define setup_blocks_store_shaded_textured(swizzling, dithering, target, \
+ edge_type) \
+{ \
+ vec_8x16u u_whole; \
+ vec_8x16u v_whole; \
+ vec_8x16u r_whole; \
+ vec_8x16u g_whole; \
+ vec_8x16u b_whole; \
+ vec_4x16u u_whole_lo, u_whole_hi; \
+ vec_4x16u v_whole_lo, v_whole_hi; \
+ vec_4x16u r_whole_lo, r_whole_hi; \
+ vec_4x16u g_whole_lo, g_whole_hi; \
+ vec_4x16u b_whole_lo, b_whole_hi; \
+ \
+ vec_8x8u u; \
+ vec_8x8u v; \
+ vec_8x8u r; \
+ vec_8x8u g; \
+ vec_8x8u b; \
+ \
+ vec_4x32u dx4; \
+ vec_4x32u dx8; \
+ \
+ gvshrn_n_u32(u_whole_lo, u_block, 16); \
+ gvshrn_n_u32(v_whole_lo, v_block, 16); \
+ gvshrn_n_u32(r_whole_lo, r_block, 16); \
+ gvshrn_n_u32(g_whole_lo, g_block, 16); \
+ gvshrn_n_u32(b_whole_lo, b_block, 16); \
+ \
+ gvdupq_l_u32(dx4, gvlo(uvrg_dx4), 0); \
+ gvaddhn_u32(u_whole_hi, u_block, dx4); \
+ gvdupq_l_u32(dx4, gvlo(uvrg_dx4), 1); \
+ gvaddhn_u32(v_whole_hi, v_block, dx4); \
+ gvdupq_l_u32(dx4, gvhi(uvrg_dx4), 0); \
+ gvaddhn_u32(r_whole_hi, r_block, dx4); \
+ gvdupq_l_u32(dx4, gvhi(uvrg_dx4), 1); \
+ gvaddhn_u32(g_whole_hi, g_block, dx4); \
+ gvdupq_n_u32(dx4, b_dx4); \
+ gvaddhn_u32(b_whole_hi, b_block, dx4); \
+ \
+ gvcombine_u16(u_whole, u_whole_lo, u_whole_hi); \
+ gvcombine_u16(v_whole, v_whole_lo, v_whole_hi); \
+ gvcombine_u16(r_whole, r_whole_lo, r_whole_hi); \
+ gvcombine_u16(g_whole, g_whole_lo, g_whole_hi); \
+ gvcombine_u16(b_whole, b_whole_lo, b_whole_hi); \
+ gvmovn_u16(u, u_whole); \
+ gvmovn_u16(v, v_whole); \
+ gvmovn_u16(r, r_whole); \
+ gvmovn_u16(g, g_whole); \
+ gvmovn_u16(b, b_whole); \
+ \
+ gvdupq_l_u32(dx8, gvlo(uvrg_dx8), 0); \
+ gvaddq_u32(u_block, u_block, dx8); \
+ gvdupq_l_u32(dx8, gvlo(uvrg_dx8), 1); \
+ gvaddq_u32(v_block, v_block, dx8); \
+ gvdupq_l_u32(dx8, gvhi(uvrg_dx8), 0); \
+ gvaddq_u32(r_block, r_block, dx8); \
+ gvdupq_l_u32(dx8, gvhi(uvrg_dx8), 1); \
+ gvaddq_u32(g_block, g_block, dx8); \
+ gvdupq_n_u32(dx8, b_dx8); \
+ gvaddq_u32(b_block, b_block, dx8); \
+ \
+ gvand(u, u, gvlo(texture_mask)); \
+ gvand(v, v, gvhi(texture_mask)); \
+ setup_blocks_texture_##swizzling(); \
+ \
+ gvst2_u8(u, v, (u8 *)block->uv.e); \
+ gvst1_u8(r, block->r.e); \
+ gvst1_u8(g, block->g.e); \
+ gvst1_u8(b, block->b.e); \
+ gvst1q_u16(dither_offsets, (u16 *)block->dither_offsets.e); \
+ block->fb_ptr = fb_ptr; \
+} \
+
+#define setup_blocks_store_unshaded_textured(swizzling, dithering, target, \
+ edge_type) \
+{ \
+ vec_8x16u u_whole; \
+ vec_8x16u v_whole; \
+ vec_4x16u u_whole_lo, u_whole_hi; \
+ vec_4x16u v_whole_lo, v_whole_hi; \
+ \
+ vec_8x8u u; \
+ vec_8x8u v; \
+ \
+ vec_4x32u dx4; \
+ vec_4x32u dx8; \
+ \
+ gvshrn_n_u32(u_whole_lo, u_block, 16); \
+ gvshrn_n_u32(v_whole_lo, v_block, 16); \
+ \
+ gvdupq_l_u32(dx4, uv_dx4, 0); \
+ gvaddhn_u32(u_whole_hi, u_block, dx4); \
+ gvdupq_l_u32(dx4, uv_dx4, 1); \
+ gvaddhn_u32(v_whole_hi, v_block, dx4); \
+ \
+ gvcombine_u16(u_whole, u_whole_lo, u_whole_hi); \
+ gvcombine_u16(v_whole, v_whole_lo, v_whole_hi); \
+ gvmovn_u16(u, u_whole); \
+ gvmovn_u16(v, v_whole); \
+ \
+ gvdupq_l_u32(dx8, uv_dx8, 0); \
+ gvaddq_u32(u_block, u_block, dx8); \
+ gvdupq_l_u32(dx8, uv_dx8, 1); \
+ gvaddq_u32(v_block, v_block, dx8); \
+ \
+ gvand(u, u, gvlo(texture_mask)); \
+ gvand(v, v, gvhi(texture_mask)); \
+ setup_blocks_texture_##swizzling(); \
+ \
+ gvst2_u8(u, v, (u8 *)block->uv.e); \
+ gvst1q_u16(dither_offsets, (u16 *)block->dither_offsets.e); \
+ block->fb_ptr = fb_ptr; \
+} \
+
+#define setup_blocks_store_shaded_untextured_dithered() \
+ gvqadd_u8(r, r, dither_offsets); \
+ gvqadd_u8(g, g, dither_offsets); \
+ gvqadd_u8(b, b, dither_offsets); \
+ \
+ gvqsub_u8(r, r, d64_4); \
+ gvqsub_u8(g, g, d64_4); \
+ gvqsub_u8(b, b, d64_4) \
+
+#define setup_blocks_store_shaded_untextured_undithered() \
+
+#define setup_blocks_store_untextured_pixels_indirect_full(_pixels) \
+ gvst1q_u16(_pixels, block->pixels.e); \
+ block->fb_ptr = fb_ptr \
+
+#define setup_blocks_store_untextured_pixels_indirect_edge(_pixels) \
+ gvst1q_u16(_pixels, block->pixels.e); \
+ block->fb_ptr = fb_ptr \
+
+#define setup_blocks_store_shaded_untextured_seed_pixels_indirect() \
+ gvmull_u8(pixels, r, d64_1) \
+
+#define setup_blocks_store_untextured_pixels_direct_full(_pixels) \
+ gvst1q_u16(_pixels, fb_ptr) \
+
+#define setup_blocks_store_untextured_pixels_direct_edge(_pixels) \
+{ \
+ vec_8x16u fb_pixels; \
+ vec_8x16u draw_mask; \
+ vec_8x16u test_mask; \
+ \
+ gvld1q_u16(test_mask, psx_gpu->test_mask.e); \
+ gvld1q_u16(fb_pixels, fb_ptr); \
+ gvdupq_n_u16(draw_mask, span_edge_data->right_mask); \
+ gvtstq_u16(draw_mask, draw_mask, test_mask); \
+ gvbifq(fb_pixels, _pixels, draw_mask); \
+ gvst1q_u16(fb_pixels, fb_ptr); \
+} \
+
+#define setup_blocks_store_shaded_untextured_seed_pixels_direct() \
+ pixels = msb_mask; \
+ gvmlal_u8(pixels, r, d64_1) \
+
+#define setup_blocks_store_shaded_untextured(swizzling, dithering, target, \
+ edge_type) \
+{ \
+ vec_8x16u r_whole; \
+ vec_8x16u g_whole; \
+ vec_8x16u b_whole; \
+ vec_4x16u r_whole_lo, r_whole_hi; \
+ vec_4x16u g_whole_lo, g_whole_hi; \
+ vec_4x16u b_whole_lo, b_whole_hi; \
+ \
+ vec_8x8u r; \
+ vec_8x8u g; \
+ vec_8x8u b; \
+ \
+ vec_4x32u dx4; \
+ vec_4x32u dx8; \
+ \
+ vec_8x16u pixels; \
+ \
+ gvshrn_n_u32(r_whole_lo, r_block, 16); \
+ gvshrn_n_u32(g_whole_lo, g_block, 16); \
+ gvshrn_n_u32(b_whole_lo, b_block, 16); \
+ \
+ gvdupq_l_u32(dx4, gvlo(rgb_dx4), 0); \
+ gvaddhn_u32(r_whole_hi, r_block, dx4); \
+ gvdupq_l_u32(dx4, gvlo(rgb_dx4), 1); \
+ gvaddhn_u32(g_whole_hi, g_block, dx4); \
+ gvdupq_l_u32(dx4, gvhi(rgb_dx4), 0); \
+ gvaddhn_u32(b_whole_hi, b_block, dx4); \
+ \
+ gvcombine_u16(r_whole, r_whole_lo, r_whole_hi); \
+ gvcombine_u16(g_whole, g_whole_lo, g_whole_hi); \
+ gvcombine_u16(b_whole, b_whole_lo, b_whole_hi); \
+ gvmovn_u16(r, r_whole); \
+ gvmovn_u16(g, g_whole); \
+ gvmovn_u16(b, b_whole); \
+ \
+ gvdupq_l_u32(dx8, gvlo(rgb_dx8), 0); \
+ gvaddq_u32(r_block, r_block, dx8); \
+ gvdupq_l_u32(dx8, gvlo(rgb_dx8), 1); \
+ gvaddq_u32(g_block, g_block, dx8); \
+ gvdupq_l_u32(dx8, gvhi(rgb_dx8), 0); \
+ gvaddq_u32(b_block, b_block, dx8); \
+ \
+ setup_blocks_store_shaded_untextured_##dithering(); \
+ \
+ gvshr_n_u8(r, r, 3); \
+ gvbic(g, g, d64_0x07); \
+ gvbic(b, b, d64_0x07); \
+ \
+ setup_blocks_store_shaded_untextured_seed_pixels_##target(); \
+ gvmlal_u8(pixels, g, d64_4); \
+ gvmlal_u8(pixels, b, d64_128); \
+ \
+ setup_blocks_store_untextured_pixels_##target##_##edge_type(pixels); \
+} \
+
+#define setup_blocks_store_unshaded_untextured(swizzling, dithering, target, \
+ edge_type) \
+ setup_blocks_store_untextured_pixels_##target##_##edge_type(colors) \
+
+#define setup_blocks_store_draw_mask_textured_indirect(_block, bits) \
+ (_block)->draw_mask_bits = bits \
+
+#define setup_blocks_store_draw_mask_untextured_indirect(_block, bits) \
+{ \
+ vec_8x16u bits_mask; \
+ vec_8x16u test_mask; \
+ \
+ gvld1q_u16(test_mask, psx_gpu->test_mask.e); \
+ gvdupq_n_u16(bits_mask, bits); \
+ gvtstq_u16(bits_mask, bits_mask, test_mask); \
+ gvst1q_u16(bits_mask, (_block)->draw_mask.e); \
+} \
+
+#define setup_blocks_store_draw_mask_untextured_direct(_block, bits) \
+
+#define setup_blocks_add_blocks_indirect() \
+ num_blocks += span_num_blocks; \
+ \
+ if(num_blocks > MAX_BLOCKS) \
+ { \
+ psx_gpu->num_blocks = num_blocks - span_num_blocks; \
+ flush_render_block_buffer(psx_gpu); \
+ num_blocks = span_num_blocks; \
+ block = psx_gpu->blocks; \
+ } \
+
+#define setup_blocks_add_blocks_direct() \
+
+#define setup_blocks_do(shading, texturing, dithering, sw, target) \
+ setup_blocks_load_msb_mask_##target(); \
+ setup_blocks_variables_##shading##_##texturing(target); \
+ \
+ edge_data_struct *span_edge_data = psx_gpu->span_edge_data; \
+ vec_4x32u *span_uvrg_offset = (vec_4x32u *)psx_gpu->span_uvrg_offset; \
+ u32 *span_b_offset = psx_gpu->span_b_offset; \
+ \
+ block_struct *block = psx_gpu->blocks + psx_gpu->num_blocks; \
+ \
+ u32 num_spans = psx_gpu->num_spans; \
+ \
+ u16 *fb_ptr; \
+ u32 y; \
+ \
+ u32 num_blocks = psx_gpu->num_blocks; \
+ u32 span_num_blocks; \
+ \
+ while(num_spans) \
+ { \
+ span_num_blocks = span_edge_data->num_blocks; \
+ if(span_num_blocks) \
+ { \
+ y = span_edge_data->y; \
+ fb_ptr = psx_gpu->vram_out_ptr + span_edge_data->left_x + (y * 1024); \
+ \
+ setup_blocks_span_initialize_##shading##_##texturing(); \
+ setup_blocks_span_initialize_##dithering(texturing); \
+ \
+ setup_blocks_add_blocks_##target(); \
+ \
+ s32 pixel_span = span_num_blocks * 8; \
+ pixel_span -= __builtin_popcount(span_edge_data->right_mask & 0xFF); \
+ \
+ span_num_blocks--; \
+ while(span_num_blocks) \
+ { \
+ setup_blocks_store_##shading##_##texturing(sw, dithering, target, \
+ full); \
+ setup_blocks_store_draw_mask_##texturing##_##target(block, 0x00); \
+ \
+ fb_ptr += 8; \
+ block++; \
+ span_num_blocks--; \
+ } \
+ \
+ setup_blocks_store_##shading##_##texturing(sw, dithering, target, edge); \
+ setup_blocks_store_draw_mask_##texturing##_##target(block, \
+ span_edge_data->right_mask); \
+ \
+ block++; \
+ } \
+ \
+ num_spans--; \
+ span_edge_data++; \
+ span_uvrg_offset++; \
+ span_b_offset++; \
+ } \
+ \
+ psx_gpu->num_blocks = num_blocks \
+
+void setup_blocks_shaded_textured_dithered_swizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_textured_dithered_swizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, textured, dithered, swizzled, indirect);
+}
+
+void setup_blocks_shaded_textured_dithered_unswizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_textured_dithered_unswizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, textured, dithered, unswizzled, indirect);
+}
+
+void setup_blocks_unshaded_textured_dithered_swizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_unshaded_textured_dithered_swizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(unshaded, textured, dithered, swizzled, indirect);
+}
+
+void setup_blocks_unshaded_textured_dithered_unswizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_unshaded_textured_dithered_unswizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(unshaded, textured, dithered, unswizzled, indirect);
+}
+
+void setup_blocks_unshaded_untextured_undithered_unswizzled_indirect(
+ psx_gpu_struct *psx_gpu)
+{
+#if 0
+ setup_blocks_unshaded_untextured_undithered_unswizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(unshaded, untextured, undithered, unswizzled, indirect);
+}
+
+void setup_blocks_unshaded_untextured_undithered_unswizzled_direct(
+ psx_gpu_struct *psx_gpu)
+{
+#if 0
+ setup_blocks_unshaded_untextured_undithered_unswizzled_direct_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(unshaded, untextured, undithered, unswizzled, direct);
+}
+
+void setup_blocks_shaded_untextured_undithered_unswizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_untextured_undithered_unswizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, untextured, undithered, unswizzled, indirect);
+}
+
+void setup_blocks_shaded_untextured_dithered_unswizzled_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_untextured_dithered_unswizzled_indirect_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, untextured, dithered, unswizzled, indirect);
+}
+
+void setup_blocks_shaded_untextured_undithered_unswizzled_direct(
+ psx_gpu_struct *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_untextured_undithered_unswizzled_direct_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, untextured, undithered, unswizzled, direct);
+}
+
+void setup_blocks_shaded_untextured_dithered_unswizzled_direct(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ setup_blocks_shaded_untextured_dithered_unswizzled_direct_(psx_gpu);
+ return;
+#endif
+ setup_blocks_do(shaded, untextured, dithered, unswizzled, direct);
+}
+
+static void update_texture_4bpp_cache(psx_gpu_struct *psx_gpu)
+{
+ u32 current_texture_page = psx_gpu->current_texture_page;
+ u8 *texture_page_ptr = psx_gpu->texture_page_base;
+ const u16 *vram_ptr = psx_gpu->vram_ptr;
+ u32 tile_x, tile_y;
+ u32 sub_y;
+ vec_8x16u c_0x00f0;
+
+ vram_ptr += (current_texture_page >> 4) * 256 * 1024;
+ vram_ptr += (current_texture_page & 0xF) * 64;
+
+ gvdupq_n_u16(c_0x00f0, 0x00f0);
+
+ psx_gpu->dirty_textures_4bpp_mask &= ~(psx_gpu->current_texture_mask);
+
+ for (tile_y = 16; tile_y; tile_y--)
+ {
+ for (tile_x = 16; tile_x; tile_x--)
+ {
+ for (sub_y = 8; sub_y; sub_y--)
+ {
+ vec_8x8u texel_block_a, texel_block_b;
+ vec_8x16u texel_block_expanded_a, texel_block_expanded_b;
+ vec_8x16u texel_block_expanded_c, texel_block_expanded_d;
+ vec_8x16u texel_block_expanded_ab, texel_block_expanded_cd;
+
+ gvld1_u8(texel_block_a, (u8 *)vram_ptr); vram_ptr += 1024;
+ gvld1_u8(texel_block_b, (u8 *)vram_ptr); vram_ptr += 1024;
+
+ gvmovl_u8(texel_block_expanded_a, texel_block_a);
+ gvshll_n_u8(texel_block_expanded_b, texel_block_a, 4);
+ gvmovl_u8(texel_block_expanded_c, texel_block_b);
+ gvshll_n_u8(texel_block_expanded_d, texel_block_b, 4);
+
+ gvbicq(texel_block_expanded_a, texel_block_expanded_a, c_0x00f0);
+ gvbicq(texel_block_expanded_b, texel_block_expanded_b, c_0x00f0);
+ gvbicq(texel_block_expanded_c, texel_block_expanded_c, c_0x00f0);
+ gvbicq(texel_block_expanded_d, texel_block_expanded_d, c_0x00f0);
+
+ gvorrq(texel_block_expanded_ab, texel_block_expanded_a, texel_block_expanded_b);
+ gvorrq(texel_block_expanded_cd, texel_block_expanded_c, texel_block_expanded_d);
+
+ gvst1q_2_pi_u32(texel_block_expanded_ab, texel_block_expanded_cd, texture_page_ptr);
+ }
+
+ vram_ptr -= (1024 * 16) - 4;
+ }
+
+ vram_ptr += (16 * 1024) - (4 * 16);
+ }
+}
+
+void update_texture_8bpp_cache_slice(psx_gpu_struct *psx_gpu,
+ u32 texture_page)
+{
+#if 0
+ update_texture_8bpp_cache_slice_(psx_gpu, texture_page);
+ return;
+#endif
+ u16 *texture_page_ptr = psx_gpu->texture_page_base;
+ u16 *vram_ptr = psx_gpu->vram_ptr;
+
+ u32 tile_x, tile_y;
+ u32 sub_y;
+
+ vram_ptr += (texture_page >> 4) * 256 * 1024;
+ vram_ptr += (texture_page & 0xF) * 64;
+
+ if((texture_page ^ psx_gpu->current_texture_page) & 0x1)
+ texture_page_ptr += (8 * 16) * 8;
+
+ for (tile_y = 16; tile_y; tile_y--)
+ {
+ for (tile_x = 8; tile_x; tile_x--)
+ {
+ for (sub_y = 4; sub_y; sub_y--)
+ {
+ vec_4x32u texels_a, texels_b, texels_c, texels_d = {};
+ gvld1q_u32(texels_a, vram_ptr); vram_ptr += 1024;
+ gvld1q_u32(texels_b, vram_ptr); vram_ptr += 1024;
+ gvld1q_u32(texels_c, vram_ptr); vram_ptr += 1024;
+ gvld1q_u32(texels_d, vram_ptr); vram_ptr += 1024;
+
+ gvst1q_2_pi_u32(texels_a, texels_b, texture_page_ptr);
+ gvst1q_2_pi_u32(texels_c, texels_d, texture_page_ptr);
+ }
+
+ vram_ptr -= (1024 * 16) - 8;
+ }
+
+ vram_ptr -= (8 * 8);
+ vram_ptr += (16 * 1024);
+
+ texture_page_ptr += (8 * 16) * 8;
+ }
+}
+
+void texture_blocks_untextured(psx_gpu_struct *psx_gpu)
+{
+}
+
+void texture_blocks_4bpp(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ texture_blocks_4bpp_(psx_gpu);
+ return;
+#endif
+ block_struct *block = psx_gpu->blocks;
+ u32 num_blocks = psx_gpu->num_blocks;
+
+ vec_8x8u texels_low;
+ vec_8x8u texels_high;
+
+ vec_16x8u clut_low;
+ vec_16x8u clut_high;
+
+ const u8 *texture_ptr_8bpp = psx_gpu->texture_page_ptr;
+
+ gvld2q_u8(clut_low, clut_high, (u8 *)psx_gpu->clut_ptr);
+
+ if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_4bpp_mask)
+ update_texture_4bpp_cache(psx_gpu);
+
+ while(num_blocks)
+ {
+ vec_8x8u texels =
+ {
+ .u8 =
+ {
+ texture_ptr_8bpp[block->uv.e[0]],
+ texture_ptr_8bpp[block->uv.e[1]],
+ texture_ptr_8bpp[block->uv.e[2]],
+ texture_ptr_8bpp[block->uv.e[3]],
+ texture_ptr_8bpp[block->uv.e[4]],
+ texture_ptr_8bpp[block->uv.e[5]],
+ texture_ptr_8bpp[block->uv.e[6]],
+ texture_ptr_8bpp[block->uv.e[7]]
+ }
+ };
+
+ gvtbl2_u8(texels_low, clut_low, texels);
+ gvtbl2_u8(texels_high, clut_high, texels);
+
+ gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e);
+
+ num_blocks--;
+ block++;
+ }
+}
+
+void texture_blocks_8bpp(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ texture_blocks_8bpp_(psx_gpu);
+ return;
+#endif
+ u32 num_blocks = psx_gpu->num_blocks;
+
+ if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_8bpp_mask)
+ update_texture_8bpp_cache(psx_gpu);
+
+ const u8 * __restrict__ texture_ptr_8bpp = psx_gpu->texture_page_ptr;
+ const u16 * __restrict__ clut_ptr = psx_gpu->clut_ptr;
+ block_struct * __restrict__ block = psx_gpu->blocks;
+
+ while(num_blocks)
+ {
+ u16 offset;
+ #define load_one(i_) \
+ offset = block->uv.e[i_]; u16 texel##i_ = texture_ptr_8bpp[offset]
+ #define store_one(i_) \
+ block->texels.e[i_] = clut_ptr[texel##i_]
+ load_one(0); load_one(1); load_one(2); load_one(3);
+ load_one(4); load_one(5); load_one(6); load_one(7);
+ store_one(0); store_one(1); store_one(2); store_one(3);
+ store_one(4); store_one(5); store_one(6); store_one(7);
+ #undef load_one
+ #undef store_one
+
+ num_blocks--;
+ block++;
+ }
+}
+
+void texture_blocks_16bpp(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ texture_blocks_16bpp_(psx_gpu);
+ return;
+#endif
+ u32 num_blocks = psx_gpu->num_blocks;
+ const u16 * __restrict__ texture_ptr_16bpp = psx_gpu->texture_page_ptr;
+ block_struct * __restrict__ block = psx_gpu->blocks;
+
+ while(num_blocks)
+ {
+ u32 offset;
+ #define load_one(i_) \
+ offset = block->uv.e[i_]; \
+ offset += ((offset & 0xFF00) * 3); \
+ u16 texel##i_ = texture_ptr_16bpp[offset]
+ #define store_one(i_) \
+ block->texels.e[i_] = texel##i_
+ load_one(0); load_one(1); load_one(2); load_one(3);
+ load_one(4); load_one(5); load_one(6); load_one(7);
+ store_one(0); store_one(1); store_one(2); store_one(3);
+ store_one(4); store_one(5); store_one(6); store_one(7);
+ #undef load_one
+ #undef store_one
+
+ num_blocks--;
+ block++;
+ }
+}
+
+#define shade_blocks_load_msb_mask_indirect() \
+
+#define shade_blocks_load_msb_mask_direct() \
+ vec_8x16u msb_mask; \
+ gvdupq_n_u16(msb_mask, psx_gpu->mask_msb); \
+
+#define shade_blocks_store_indirect(_draw_mask, _pixels) \
+ gvst1q_u16(_draw_mask, block->draw_mask.e); \
+ gvst1q_u16(_pixels, block->pixels.e); \
+
+#define shade_blocks_store_direct(_draw_mask, _pixels) \
+{ \
+ vec_8x16u fb_pixels; \
+ gvorrq(_pixels, _pixels, msb_mask); \
+ gvld1q_u16(fb_pixels, block->fb_ptr); \
+ gvbifq(fb_pixels, _pixels, _draw_mask); \
+ gvst1q_u16(fb_pixels, block->fb_ptr); \
+} \
+
+#define shade_blocks_textured_false_modulated_check_dithered(target) \
+
+#define shade_blocks_textured_false_modulated_check_undithered(target) \
+ if(psx_gpu->triangle_color == 0x808080) \
+ { \
+ shade_blocks_textured_unmodulated_##target(psx_gpu); \
+ return; \
+ } \
+
+#define shade_blocks_textured_modulated_shaded_primitive_load(dithering, \
+ target) \
+
+#define shade_blocks_textured_modulated_unshaded_primitive_load(dithering, \
+ target) \
+{ \
+ u32 color = psx_gpu->triangle_color; \
+ gvdup_n_u8(colors_r, color); \
+ gvdup_n_u8(colors_g, color >> 8); \
+ gvdup_n_u8(colors_b, color >> 16); \
+ shade_blocks_textured_false_modulated_check_##dithering(target); \
+} \
+
+#define shade_blocks_textured_modulated_shaded_block_load() \
+ gvld1_u8(colors_r, block->r.e); \
+ gvld1_u8(colors_g, block->g.e); \
+ gvld1_u8(colors_b, block->b.e) \
+
+#define shade_blocks_textured_modulated_unshaded_block_load() \
+
+#define shade_blocks_textured_modulate_dithered(component) \
+ gvld1q_u16(pixels_##component, block->dither_offsets.e); \
+ gvmlal_u8(pixels_##component, texels_##component, colors_##component) \
+
+#define shade_blocks_textured_modulate_undithered(component) \
+ gvmull_u8(pixels_##component, texels_##component, colors_##component) \
+
+#define shade_blocks_textured_modulated_do(shading, dithering, target) \
+ block_struct *block = psx_gpu->blocks; \
+ u32 num_blocks = psx_gpu->num_blocks; \
+ vec_8x16u texels; \
+ \
+ vec_8x8u texels_r; \
+ vec_8x8u texels_g; \
+ vec_8x8u texels_b; \
+ \
+ vec_8x8u colors_r; \
+ vec_8x8u colors_g; \
+ vec_8x8u colors_b; \
+ \
+ vec_8x8u pixels_r_low; \
+ vec_8x8u pixels_g_low; \
+ vec_8x8u pixels_b_low; \
+ vec_8x16u pixels; \
+ \
+ vec_8x16u pixels_r; \
+ vec_8x16u pixels_g; \
+ vec_8x16u pixels_b; \
+ \
+ vec_8x16u draw_mask; \
+ vec_8x16u zero_mask; \
+ \
+ vec_8x8u d64_0x07; \
+ vec_8x8u d64_0x1F; \
+ vec_8x8u d64_1; \
+ vec_8x8u d64_4; \
+ vec_8x8u d64_128; \
+ \
+ vec_8x16u d128_0x8000; \
+ \
+ vec_8x16u test_mask; \
+ u32 draw_mask_bits; \
+ \
+ gvld1q_u16(test_mask, psx_gpu->test_mask.e); \
+ shade_blocks_load_msb_mask_##target(); \
+ \
+ gvdup_n_u8(d64_0x07, 0x07); \
+ gvdup_n_u8(d64_0x1F, 0x1F); \
+ gvdup_n_u8(d64_1, 1); \
+ gvdup_n_u8(d64_4, 4); \
+ gvdup_n_u8(d64_128, 128); \
+ \
+ gvdupq_n_u16(d128_0x8000, 0x8000); \
+ \
+ shade_blocks_textured_modulated_##shading##_primitive_load(dithering, \
+ target); \
+ \
+ while(num_blocks) \
+ { \
+ draw_mask_bits = block->draw_mask_bits; \
+ gvdupq_n_u16(draw_mask, draw_mask_bits); \
+ gvtstq_u16(draw_mask, draw_mask, test_mask); \
+ \
+ shade_blocks_textured_modulated_##shading##_block_load(); \
+ \
+ gvld1q_u16(texels, block->texels.e); \
+ \
+ gvmovn_u16(texels_r, texels); \
+ gvshrn_n_u16(texels_g, texels, 5); \
+ gvshrn_n_u16(texels_b, texels, 7); \
+ \
+ gvand(texels_r, texels_r, d64_0x1F); \
+ gvand(texels_g, texels_g, d64_0x1F); \
+ gvshr_n_u8(texels_b, texels_b, 3); \
+ \
+ shade_blocks_textured_modulate_##dithering(r); \
+ shade_blocks_textured_modulate_##dithering(g); \
+ shade_blocks_textured_modulate_##dithering(b); \
+ \
+ gvceqzq_u16(zero_mask, texels); \
+ gvand(pixels, texels, d128_0x8000); \
+ \
+ gvqshrun_n_s16(pixels_r_low, pixels_r, 4); \
+ gvqshrun_n_s16(pixels_g_low, pixels_g, 4); \
+ gvqshrun_n_s16(pixels_b_low, pixels_b, 4); \
+ \
+ gvorrq(zero_mask, draw_mask, zero_mask); \
+ \
+ gvshr_n_u8(pixels_r_low, pixels_r_low, 3); \
+ gvbic(pixels_g_low, pixels_g_low, d64_0x07); \
+ gvbic(pixels_b_low, pixels_b_low, d64_0x07); \
+ \
+ gvmlal_u8(pixels, pixels_r_low, d64_1); \
+ gvmlal_u8(pixels, pixels_g_low, d64_4); \
+ gvmlal_u8(pixels, pixels_b_low, d64_128); \
+ \
+ shade_blocks_store_##target(zero_mask, pixels); \
+ \
+ num_blocks--; \
+ block++; \
+ } \
+
+void shade_blocks_shaded_textured_modulated_dithered_direct(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_shaded_textured_modulated_dithered_direct_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(shaded, dithered, direct);
+}
+
+void shade_blocks_shaded_textured_modulated_undithered_direct(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_shaded_textured_modulated_undithered_direct_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(shaded, undithered, direct);
+}
+
+void shade_blocks_unshaded_textured_modulated_dithered_direct(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_unshaded_textured_modulated_dithered_direct_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(unshaded, dithered, direct);
+}
+
+void shade_blocks_unshaded_textured_modulated_undithered_direct(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_unshaded_textured_modulated_undithered_direct_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(unshaded, undithered, direct);
+}
+
+void shade_blocks_shaded_textured_modulated_dithered_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_shaded_textured_modulated_dithered_indirect_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(shaded, dithered, indirect);
+}
+
+void shade_blocks_shaded_textured_modulated_undithered_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_shaded_textured_modulated_undithered_indirect_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(shaded, undithered, indirect);
+}
+
+void shade_blocks_unshaded_textured_modulated_dithered_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_unshaded_textured_modulated_dithered_indirect_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(unshaded, dithered, indirect);
+}
+
+void shade_blocks_unshaded_textured_modulated_undithered_indirect(psx_gpu_struct
+ *psx_gpu)
+{
+#if 0
+ shade_blocks_unshaded_textured_modulated_undithered_indirect_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_modulated_do(unshaded, undithered, indirect);
+}
+
+#define shade_blocks_textured_unmodulated_do(target) \
+ block_struct *block = psx_gpu->blocks; \
+ u32 num_blocks = psx_gpu->num_blocks; \
+ vec_8x16u draw_mask; \
+ vec_8x16u test_mask; \
+ u32 draw_mask_bits; \
+ \
+ vec_8x16u pixels; \
+ \
+ gvld1q_u16(test_mask, psx_gpu->test_mask.e); \
+ shade_blocks_load_msb_mask_##target(); \
+ \
+ while(num_blocks) \
+ { \
+ vec_8x16u zero_mask; \
+ \
+ draw_mask_bits = block->draw_mask_bits; \
+ gvdupq_n_u16(draw_mask, draw_mask_bits); \
+ gvtstq_u16(draw_mask, draw_mask, test_mask); \
+ \
+ gvld1q_u16(pixels, block->texels.e); \
+ \
+ gvceqzq_u16(zero_mask, pixels); \
+ gvorrq(zero_mask, draw_mask, zero_mask); \
+ \
+ shade_blocks_store_##target(zero_mask, pixels); \
+ \
+ num_blocks--; \
+ block++; \
+ } \
+
+void shade_blocks_textured_unmodulated_indirect(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ shade_blocks_textured_unmodulated_indirect_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_unmodulated_do(indirect)
+}
+
+void shade_blocks_textured_unmodulated_direct(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ shade_blocks_textured_unmodulated_direct_(psx_gpu);
+ return;
+#endif
+ shade_blocks_textured_unmodulated_do(direct)
+}
+
+void shade_blocks_unshaded_untextured_indirect(psx_gpu_struct *psx_gpu)
+{
+}
+
+void shade_blocks_unshaded_untextured_direct(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ shade_blocks_unshaded_untextured_direct_(psx_gpu);
+ return;
+#endif
+ block_struct *block = psx_gpu->blocks;
+ u32 num_blocks = psx_gpu->num_blocks;
+
+ vec_8x16u pixels;
+ gvld1q_u16(pixels, block->texels.e);
+ shade_blocks_load_msb_mask_direct();
+
+ while(num_blocks)
+ {
+ vec_8x16u draw_mask;
+ gvld1q_u16(draw_mask, block->draw_mask.e);
+ shade_blocks_store_direct(draw_mask, pixels);
+
+ num_blocks--;
+ block++;
+ }
+}
+
+#define blend_blocks_mask_evaluate_on() \
+ vec_8x16u mask_pixels; \
+ gvcltzq_s16(mask_pixels, framebuffer_pixels); \
+ gvorrq(draw_mask, draw_mask, mask_pixels) \
+
+#define blend_blocks_mask_evaluate_off() \
+
+#define blend_blocks_average() \
+{ \
+ vec_8x16u pixels_no_msb; \
+ vec_8x16u fb_pixels_no_msb; \
+ \
+ vec_8x16u d128_0x0421; \
+ \
+ gvdupq_n_u16(d128_0x0421, 0x0421); \
+ \
+ gveorq(blend_pixels, pixels, framebuffer_pixels); \
+ gvbicq(pixels_no_msb, pixels, d128_0x8000); \
+ gvand(blend_pixels, blend_pixels, d128_0x0421); \
+ gvsubq_u16(blend_pixels, pixels_no_msb, blend_pixels); \
+ gvbicq(fb_pixels_no_msb, framebuffer_pixels, d128_0x8000); \
+ gvhaddq_u16(blend_pixels, fb_pixels_no_msb, blend_pixels); \
+} \
+
+#define blend_blocks_add() \
+{ \
+ vec_8x16u pixels_rb, pixels_g; \
+ vec_8x16u fb_rb, fb_g; \
+ \
+ vec_8x16u d128_0x7C1F; \
+ vec_8x16u d128_0x03E0; \
+ \
+ gvdupq_n_u16(d128_0x7C1F, 0x7C1F); \
+ gvdupq_n_u16(d128_0x03E0, 0x03E0); \
+ \
+ gvand(pixels_rb, pixels, d128_0x7C1F); \
+ gvand(pixels_g, pixels, d128_0x03E0); \
+ \
+ gvand(fb_rb, framebuffer_pixels, d128_0x7C1F); \
+ gvand(fb_g, framebuffer_pixels, d128_0x03E0); \
+ \
+ gvaddq_u16(fb_rb, fb_rb, pixels_rb); \
+ gvaddq_u16(fb_g, fb_g, pixels_g); \
+ \
+ gvminq_u8(fb_rb, fb_rb, d128_0x7C1F); \
+ gvminq_u16(fb_g, fb_g, d128_0x03E0); \
+ \
+ gvorrq(blend_pixels, fb_rb, fb_g); \
+} \
+
+#define blend_blocks_subtract() \
+{ \
+ vec_8x16u pixels_rb, pixels_g; \
+ vec_8x16u fb_rb, fb_g; \
+ \
+ vec_8x16u d128_0x7C1F; \
+ vec_8x16u d128_0x03E0; \
+ \
+ gvdupq_n_u16(d128_0x7C1F, 0x7C1F); \
+ gvdupq_n_u16(d128_0x03E0, 0x03E0); \
+ \
+ gvand(pixels_rb, pixels, d128_0x7C1F); \
+ gvand(pixels_g, pixels, d128_0x03E0); \
+ \
+ gvand(fb_rb, framebuffer_pixels, d128_0x7C1F); \
+ gvand(fb_g, framebuffer_pixels, d128_0x03E0); \
+ \
+ gvqsubq_u8(fb_rb, fb_rb, pixels_rb); \
+ gvqsubq_u16(fb_g, fb_g, pixels_g); \
+ \
+ gvorrq(blend_pixels, fb_rb, fb_g); \
+} \
+
+#define blend_blocks_add_fourth() \
+{ \
+ vec_8x16u pixels_rb, pixels_g; \
+ vec_8x16u pixels_fourth; \
+ vec_8x16u fb_rb, fb_g; \
+ \
+ vec_8x16u d128_0x7C1F; \
+ vec_8x16u d128_0x1C07; \
+ vec_8x16u d128_0x03E0; \
+ vec_8x16u d128_0x00E0; \
+ \
+ gvdupq_n_u16(d128_0x7C1F, 0x7C1F); \
+ gvdupq_n_u16(d128_0x1C07, 0x1C07); \
+ gvdupq_n_u16(d128_0x03E0, 0x03E0); \
+ gvdupq_n_u16(d128_0x00E0, 0x00E0); \
+ \
+ gvshrq_n_u16(pixels_fourth, pixels, 2); \
+ \
+ gvand(fb_rb, framebuffer_pixels, d128_0x7C1F); \
+ gvand(fb_g, framebuffer_pixels, d128_0x03E0); \
+ \
+ gvand(pixels_rb, pixels_fourth, d128_0x1C07); \
+ gvand(pixels_g, pixels_fourth, d128_0x00E0); \
+ \
+ gvaddq_u16(fb_rb, fb_rb, pixels_rb); \
+ gvaddq_u16(fb_g, fb_g, pixels_g); \
+ \
+ gvminq_u8(fb_rb, fb_rb, d128_0x7C1F); \
+ gvminq_u16(fb_g, fb_g, d128_0x03E0); \
+ \
+ gvorrq(blend_pixels, fb_rb, fb_g); \
+} \
+
+#define blend_blocks_blended_combine_textured() \
+{ \
+ vec_8x16u blend_mask; \
+ gvcltzq_s16(blend_mask, pixels); \
+ \
+ gvorrq(blend_pixels, blend_pixels, d128_0x8000); \
+ gvbifq(blend_pixels, pixels, blend_mask); \
+} \
+
+#define blend_blocks_blended_combine_untextured() \
+
+#define blend_blocks_body_blend(blend_mode, texturing) \
+{ \
+ blend_blocks_##blend_mode(); \
+ blend_blocks_blended_combine_##texturing(); \
+} \
+
+#define blend_blocks_body_average(texturing) \
+ blend_blocks_body_blend(average, texturing) \
+
+#define blend_blocks_body_add(texturing) \
+ blend_blocks_body_blend(add, texturing) \
+
+#define blend_blocks_body_subtract(texturing) \
+ blend_blocks_body_blend(subtract, texturing) \
+
+#define blend_blocks_body_add_fourth(texturing) \
+ blend_blocks_body_blend(add_fourth, texturing) \
+
+#define blend_blocks_body_unblended(texturing) \
+ blend_pixels = pixels \
+
+#define blend_blocks_do(texturing, blend_mode, mask_evaluate) \
+ block_struct *block = psx_gpu->blocks; \
+ u32 num_blocks = psx_gpu->num_blocks; \
+ vec_8x16u draw_mask; \
+ vec_8x16u pixels; \
+ vec_8x16u blend_pixels; \
+ vec_8x16u framebuffer_pixels; \
+ vec_8x16u msb_mask; \
+ vec_8x16u d128_0x8000; \
+ \
+ u16 *fb_ptr; \
+ \
+ gvdupq_n_u16(d128_0x8000, 0x8000); \
+ gvdupq_n_u16(msb_mask, psx_gpu->mask_msb); \
+ (void)d128_0x8000; /* sometimes unused */ \
+ \
+ while(num_blocks) \
+ { \
+ gvld1q_u16(pixels, block->pixels.e); \
+ gvld1q_u16(draw_mask, block->draw_mask.e); \
+ fb_ptr = block->fb_ptr; \
+ \
+ gvld1q_u16(framebuffer_pixels, fb_ptr); \
+ \
+ blend_blocks_mask_evaluate_##mask_evaluate(); \
+ blend_blocks_body_##blend_mode(texturing); \
+ \
+ gvorrq(blend_pixels, blend_pixels, msb_mask); \
+ gvbifq(framebuffer_pixels, blend_pixels, draw_mask); \
+ gvst1q_u16(framebuffer_pixels, fb_ptr); \
+ \
+ num_blocks--; \
+ block++; \
+ } \
+
+
+void blend_blocks_textured_average_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_average_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, average, off);
+}
+
+void blend_blocks_untextured_average_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_average_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, average, off);
+}
+
+void blend_blocks_textured_average_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_average_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, average, on);
+}
+
+void blend_blocks_untextured_average_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_average_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, average, on);
+}
+
+void blend_blocks_textured_add_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_add_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, add, off);
+}
+
+void blend_blocks_textured_add_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_add_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, add, on);
+}
+
+void blend_blocks_untextured_add_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_add_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, add, off);
+}
+
+void blend_blocks_untextured_add_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_add_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, add, on);
+}
+
+void blend_blocks_textured_subtract_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_subtract_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, subtract, off);
+}
+
+void blend_blocks_textured_subtract_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_subtract_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, subtract, on);
+}
+
+void blend_blocks_untextured_subtract_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_subtract_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, subtract, off);
+}
+
+void blend_blocks_untextured_subtract_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_subtract_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, subtract, on);
+}
+
+void blend_blocks_textured_add_fourth_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_add_fourth_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, add_fourth, off);
+}
+
+void blend_blocks_textured_add_fourth_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_add_fourth_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, add_fourth, on);
+}
+
+void blend_blocks_untextured_add_fourth_off(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_add_fourth_off_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, add_fourth, off);
+}
+
+void blend_blocks_untextured_add_fourth_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_untextured_add_fourth_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(untextured, add_fourth, on);
+}
+
+void blend_blocks_textured_unblended_on(psx_gpu_struct *psx_gpu)
+{
+#if 0
+ blend_blocks_textured_unblended_on_(psx_gpu);
+ return;
+#endif
+ blend_blocks_do(textured, unblended, on);
+}
+
+void blend_blocks_textured_unblended_off(psx_gpu_struct *psx_gpu)
+{
+}
+
+void setup_sprite_untextured(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u,
+ s32 v, s32 width, s32 height, u32 color)
+{
+ if((psx_gpu->render_state & (RENDER_STATE_MASK_EVALUATE |
+ RENDER_FLAGS_MODULATE_TEXELS | RENDER_FLAGS_BLEND)) == 0 &&
+ (psx_gpu->render_mode & RENDER_INTERLACE_ENABLED) == 0)
+ {
+ setup_sprite_untextured_simple(psx_gpu, x, y, u, v, width, height, color);
+ return;
+ }
+
+#if 0
+ setup_sprite_untextured_(psx_gpu, x, y, u, v, width, height, color);
+ return;
+#endif
+ u32 right_width = ((width - 1) & 0x7) + 1;
+ u32 right_mask_bits = (0xFF << right_width);
+ u16 *fb_ptr = psx_gpu->vram_out_ptr + (y * 1024) + x;
+ u32 block_width = (width + 7) / 8;
+ u32 fb_ptr_pitch = 1024 - ((block_width - 1) * 8);
+ u32 blocks_remaining;
+ u32 num_blocks = psx_gpu->num_blocks;
+ block_struct *block = psx_gpu->blocks + num_blocks;
+
+ u32 color_r = color & 0xFF;
+ u32 color_g = (color >> 8) & 0xFF;
+ u32 color_b = (color >> 16) & 0xFF;
+ vec_8x16u colors;
+ vec_8x16u right_mask;
+ vec_8x16u test_mask;
+ vec_8x16u zero_mask;
+
+ gvld1q_u16(test_mask, psx_gpu->test_mask.e);
+ color = (color_r >> 3) | ((color_g >> 3) << 5) | ((color_b >> 3) << 10);
+
+ gvdupq_n_u16(colors, color);
+ gvdupq_n_u16(zero_mask, 0x00);
+ gvdupq_n_u16(right_mask, right_mask_bits);
+ gvtstq_u16(right_mask, right_mask, test_mask);
+
+ while(height)
+ {
+ blocks_remaining = block_width - 1;
+ num_blocks += block_width;
+
+ if(num_blocks > MAX_BLOCKS)
+ {
+ flush_render_block_buffer(psx_gpu);
+ num_blocks = block_width;
+ block = psx_gpu->blocks;
+ }
+
+ while(blocks_remaining)
+ {
+ gvst1q_u16(colors, block->pixels.e);
+ gvst1q_u16(zero_mask, block->draw_mask.e);
+ block->fb_ptr = fb_ptr;
+
+ fb_ptr += 8;
+ block++;
+ blocks_remaining--;
+ }
+
+ gvst1q_u16(colors, block->pixels.e);
+ gvst1q_u16(right_mask, block->draw_mask.e);
+ block->fb_ptr = fb_ptr;
+
+ block++;
+ fb_ptr += fb_ptr_pitch;
+
+ height--;
+ psx_gpu->num_blocks = num_blocks;
+ }
+}
+
+#define setup_sprite_tiled_initialize_4bpp_clut() \
+ vec_16x8u clut_low, clut_high; \
+ \
+ gvld2q_u8(clut_low, clut_high, (u8 *)psx_gpu->clut_ptr) \
+
+#define setup_sprite_tiled_initialize_4bpp() \
+ setup_sprite_tiled_initialize_4bpp_clut(); \
+ \
+ if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_4bpp_mask) \
+ update_texture_4bpp_cache(psx_gpu) \
+
+#define setup_sprite_tiled_initialize_8bpp() \
+ if(psx_gpu->current_texture_mask & psx_gpu->dirty_textures_8bpp_mask) \
+ update_texture_8bpp_cache(psx_gpu) \
+
+#define setup_sprite_tile_fetch_texel_block_8bpp(offset) \
+ texture_block_ptr = psx_gpu->texture_page_ptr + \
+ ((texture_offset + offset) & texture_mask); \
+ \
+ gvld1_u8(texels, (u8 *)texture_block_ptr) \
+
+#define setup_sprite_tile_add_blocks(tile_num_blocks) \
+ num_blocks += tile_num_blocks; \
+ \
+ if(num_blocks > MAX_BLOCKS) \
+ { \
+ flush_render_block_buffer(psx_gpu); \
+ num_blocks = tile_num_blocks; \
+ block = psx_gpu->blocks; \
+ } \
+
+#define setup_sprite_tile_full_4bpp(edge) \
+{ \
+ vec_8x8u texels_low, texels_high; \
+ setup_sprite_tile_add_blocks(sub_tile_height * 2); \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ \
+ gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e); \
+ block->draw_mask_bits = left_mask_bits; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ setup_sprite_tile_fetch_texel_block_8bpp(8); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ \
+ gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e); \
+ block->draw_mask_bits = right_mask_bits; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ fb_ptr += 1024; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_half_4bpp(edge) \
+{ \
+ vec_8x8u texels_low, texels_high; \
+ setup_sprite_tile_add_blocks(sub_tile_height); \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ \
+ gvst2_u8(texels_low, texels_high, (u8 *)block->texels.e); \
+ block->draw_mask_bits = edge##_mask_bits; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ fb_ptr += 1024; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_full_8bpp(edge) \
+{ \
+ setup_sprite_tile_add_blocks(sub_tile_height * 2); \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvst1_u8(texels, block->r.e); \
+ block->draw_mask_bits = left_mask_bits; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ setup_sprite_tile_fetch_texel_block_8bpp(8); \
+ gvst1_u8(texels, block->r.e); \
+ block->draw_mask_bits = right_mask_bits; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ fb_ptr += 1024; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_half_8bpp(edge) \
+{ \
+ setup_sprite_tile_add_blocks(sub_tile_height * 2); \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvst1_u8(texels, block->r.e); \
+ block->draw_mask_bits = edge##_mask_bits; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ fb_ptr += 1024; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half_right() \
+ texture_offset = texture_offset_base + 8; \
+ fb_ptr += 8 \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half_left() \
+ texture_offset = texture_offset_base \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half(edge) \
+ setup_sprite_tile_column_edge_pre_adjust_half_##edge() \
+
+#define setup_sprite_tile_column_edge_pre_adjust_full(edge) \
+ texture_offset = texture_offset_base \
+
+#define setup_sprite_tile_column_edge_post_adjust_half_right() \
+ fb_ptr -= 8 \
+
+#define setup_sprite_tile_column_edge_post_adjust_half_left() \
+
+#define setup_sprite_tile_column_edge_post_adjust_half(edge) \
+ setup_sprite_tile_column_edge_post_adjust_half_##edge() \
+
+#define setup_sprite_tile_column_edge_post_adjust_full(edge) \
+
+
+#define setup_sprite_tile_column_height_single(edge_mode, edge, texture_mode, \
+ x4mode) \
+do \
+{ \
+ sub_tile_height = column_data; \
+ setup_sprite_tile_column_edge_pre_adjust_##edge_mode##x4mode(edge); \
+ setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge); \
+ setup_sprite_tile_column_edge_post_adjust_##edge_mode##x4mode(edge); \
+} while(0) \
+
+#define setup_sprite_tile_column_height_multi(edge_mode, edge, texture_mode, \
+ x4mode) \
+do \
+{ \
+ u32 tiles_remaining = column_data >> 16; \
+ sub_tile_height = column_data & 0xFF; \
+ setup_sprite_tile_column_edge_pre_adjust_##edge_mode##x4mode(edge); \
+ setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge); \
+ tiles_remaining -= 1; \
+ \
+ while(tiles_remaining) \
+ { \
+ sub_tile_height = 16; \
+ setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge); \
+ tiles_remaining--; \
+ } \
+ \
+ sub_tile_height = (column_data >> 8) & 0xFF; \
+ setup_sprite_tile_##edge_mode##_##texture_mode##x4mode(edge); \
+ setup_sprite_tile_column_edge_post_adjust_##edge_mode##x4mode(edge); \
+} while(0) \
+
+
+#define setup_sprite_column_data_single() \
+ column_data = height \
+
+#define setup_sprite_column_data_multi() \
+ column_data = 16 - offset_v; \
+ column_data |= ((height_rounded & 0xF) + 1) << 8; \
+ column_data |= (tile_height - 1) << 16 \
+
+#define RIGHT_MASK_BIT_SHIFT 8
+#define RIGHT_MASK_BIT_SHIFT_4x 16
+
+#define setup_sprite_tile_column_width_single(texture_mode, multi_height, \
+ edge_mode, edge, x4mode) \
+{ \
+ setup_sprite_column_data_##multi_height(); \
+ left_mask_bits = left_block_mask | right_block_mask; \
+ right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode; \
+ \
+ setup_sprite_tile_column_height_##multi_height(edge_mode, edge, \
+ texture_mode, x4mode); \
+} \
+
+#define setup_sprite_tiled_advance_column() \
+ texture_offset_base += 0x100; \
+ if((texture_offset_base & 0xF00) == 0) \
+ texture_offset_base -= (0x100 + 0xF00) \
+
+#define FB_PTR_MULTIPLIER 1
+#define FB_PTR_MULTIPLIER_4x 2
+
+#define setup_sprite_tile_column_width_multi(texture_mode, multi_height, \
+ left_mode, right_mode, x4mode) \
+{ \
+ setup_sprite_column_data_##multi_height(); \
+ s32 fb_ptr_advance_column = (16 - (1024 * height)) \
+ * FB_PTR_MULTIPLIER##x4mode; \
+ \
+ tile_width -= 2; \
+ left_mask_bits = left_block_mask; \
+ right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode; \
+ \
+ setup_sprite_tile_column_height_##multi_height(left_mode, right, \
+ texture_mode, x4mode); \
+ fb_ptr += fb_ptr_advance_column; \
+ \
+ left_mask_bits = 0x00; \
+ right_mask_bits = 0x00; \
+ \
+ while(tile_width) \
+ { \
+ setup_sprite_tiled_advance_column(); \
+ setup_sprite_tile_column_height_##multi_height(full, none, \
+ texture_mode, x4mode); \
+ fb_ptr += fb_ptr_advance_column; \
+ tile_width--; \
+ } \
+ \
+ left_mask_bits = right_block_mask; \
+ right_mask_bits = left_mask_bits >> RIGHT_MASK_BIT_SHIFT##x4mode; \
+ \
+ setup_sprite_tiled_advance_column(); \
+ setup_sprite_tile_column_height_##multi_height(right_mode, left, \
+ texture_mode, x4mode); \
+} \
+
+
+/* 4x stuff */
+#define setup_sprite_tiled_initialize_4bpp_4x() \
+ setup_sprite_tiled_initialize_4bpp_clut() \
+
+#define setup_sprite_tiled_initialize_8bpp_4x() \
+
+#define setup_sprite_tile_full_4bpp_4x(edge) \
+{ \
+ vec_8x8u texels_low, texels_high; \
+ vec_8x16u pixels; \
+ vec_4x16u pixels_half; \
+ setup_sprite_tile_add_blocks(sub_tile_height * 2 * 4); \
+ u32 left_mask_bits_a = left_mask_bits & 0xFF; \
+ u32 left_mask_bits_b = left_mask_bits >> 8; \
+ u32 right_mask_bits_a = right_mask_bits & 0xFF; \
+ u32 right_mask_bits_b = right_mask_bits >> 8; \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ gvzip_u8(pixels, texels_low, texels_high); \
+ \
+ gvget_lo(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = left_mask_bits_a; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = left_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024; \
+ block++; \
+ \
+ gvget_hi(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = left_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = left_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 1024 + 8; \
+ block++; \
+ \
+ setup_sprite_tile_fetch_texel_block_8bpp(8); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ gvzip_u8(pixels, texels_low, texels_high); \
+ \
+ gvget_lo(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = right_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 16; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = right_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024 + 16; \
+ block++; \
+ \
+ gvget_hi(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = right_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 24; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = right_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 1024 + 24; \
+ block++; \
+ \
+ fb_ptr += 2048; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_half_4bpp_4x(edge) \
+{ \
+ vec_8x8u texels_low, texels_high; \
+ vec_8x16u pixels; \
+ vec_4x16u pixels_half; \
+ setup_sprite_tile_add_blocks(sub_tile_height * 4); \
+ u32 edge##_mask_bits_a = edge##_mask_bits & 0xFF; \
+ u32 edge##_mask_bits_b = edge##_mask_bits >> 8; \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvtbl2_u8(texels_low, clut_low, texels); \
+ gvtbl2_u8(texels_high, clut_high, texels); \
+ gvzip_u8(pixels, texels_low, texels_high); \
+ \
+ gvget_lo(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = edge##_mask_bits_a; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = edge##_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024; \
+ block++; \
+ \
+ gvget_hi(pixels_half, pixels); \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = edge##_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ gvst2_u16(pixels_half, pixels_half, block->texels.e); \
+ block->draw_mask_bits = edge##_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 1024 + 8; \
+ block++; \
+ \
+ fb_ptr += 2048; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_full_8bpp_4x(edge) \
+{ \
+ setup_sprite_tile_add_blocks(sub_tile_height * 2 * 4); \
+ vec_8x16u texels_wide; \
+ vec_4x16u texels_half; \
+ u32 left_mask_bits_a = left_mask_bits & 0xFF; \
+ u32 left_mask_bits_b = left_mask_bits >> 8; \
+ u32 right_mask_bits_a = right_mask_bits & 0xFF; \
+ u32 right_mask_bits_b = right_mask_bits >> 8; \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvzip_u8(texels_wide, texels, texels); \
+ gvget_lo(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = left_mask_bits_a; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = left_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024; \
+ block++; \
+ \
+ gvget_hi(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = left_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = left_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 1024 + 8; \
+ block++; \
+ \
+ setup_sprite_tile_fetch_texel_block_8bpp(8); \
+ gvzip_u8(texels_wide, texels, texels); \
+ gvget_lo(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = right_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 16; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = right_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024 + 16; \
+ block++; \
+ \
+ gvget_hi(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = right_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 24; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = right_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 24 + 1024; \
+ block++; \
+ \
+ fb_ptr += 2048; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_half_8bpp_4x(edge) \
+{ \
+ setup_sprite_tile_add_blocks(sub_tile_height * 4); \
+ vec_8x16u texels_wide; \
+ vec_4x16u texels_half; \
+ u32 edge##_mask_bits_a = edge##_mask_bits & 0xFF; \
+ u32 edge##_mask_bits_b = edge##_mask_bits >> 8; \
+ \
+ while(sub_tile_height) \
+ { \
+ setup_sprite_tile_fetch_texel_block_8bpp(0); \
+ gvzip_u8(texels_wide, texels, texels); \
+ gvget_lo(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = edge##_mask_bits_a; \
+ block->fb_ptr = fb_ptr; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = edge##_mask_bits_a; \
+ block->fb_ptr = fb_ptr + 1024; \
+ block++; \
+ \
+ gvget_hi(texels_half, texels_wide); \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = edge##_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 8; \
+ block++; \
+ \
+ gvst1_u8(texels_half, block->r.e); \
+ block->draw_mask_bits = edge##_mask_bits_b; \
+ block->fb_ptr = fb_ptr + 8 + 1024; \
+ block++; \
+ \
+ fb_ptr += 2048; \
+ texture_offset += 0x10; \
+ sub_tile_height--; \
+ } \
+ texture_offset += 0xF00; \
+ psx_gpu->num_blocks = num_blocks; \
+} \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half_right_4x() \
+ texture_offset = texture_offset_base + 8; \
+ fb_ptr += 16 \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half_left_4x() \
+ texture_offset = texture_offset_base \
+
+#define setup_sprite_tile_column_edge_pre_adjust_half_4x(edge) \
+ setup_sprite_tile_column_edge_pre_adjust_half_##edge##_4x() \
+
+#define setup_sprite_tile_column_edge_pre_adjust_full_4x(edge) \
+ texture_offset = texture_offset_base \
+
+#define setup_sprite_tile_column_edge_post_adjust_half_right_4x() \
+ fb_ptr -= 16 \
+
+#define setup_sprite_tile_column_edge_post_adjust_half_left_4x() \
+
+#define setup_sprite_tile_column_edge_post_adjust_half_4x(edge) \
+ setup_sprite_tile_column_edge_post_adjust_half_##edge##_4x() \
+
+#define setup_sprite_tile_column_edge_post_adjust_full_4x(edge) \
+
+#define setup_sprite_offset_u_adjust() \
+
+#define setup_sprite_comapre_left_block_mask() \
+ ((left_block_mask & 0xFF) == 0xFF) \
+
+#define setup_sprite_comapre_right_block_mask() \
+ (((right_block_mask >> 8) & 0xFF) == 0xFF) \
+
+#define setup_sprite_offset_u_adjust_4x() \
+ offset_u *= 2; \
+ offset_u_right = offset_u_right * 2 + 1 \
+
+#define setup_sprite_comapre_left_block_mask_4x() \
+ ((left_block_mask & 0xFFFF) == 0xFFFF) \
+
+#define setup_sprite_comapre_right_block_mask_4x() \
+ (((right_block_mask >> 16) & 0xFFFF) == 0xFFFF) \
+
+
+#define setup_sprite_tiled_do(texture_mode, x4mode) \
+ s32 offset_u = u & 0xF; \
+ s32 offset_v = v & 0xF; \
+ \
+ s32 width_rounded = offset_u + width + 15; \
+ s32 height_rounded = offset_v + height + 15; \
+ s32 tile_height = height_rounded / 16; \
+ s32 tile_width = width_rounded / 16; \
+ u32 offset_u_right = width_rounded & 0xF; \
+ \
+ setup_sprite_offset_u_adjust##x4mode(); \
+ \
+ u32 left_block_mask = ~(0xFFFFFFFF << offset_u); \
+ u32 right_block_mask = 0xFFFFFFFE << offset_u_right; \
+ \
+ u32 left_mask_bits; \
+ u32 right_mask_bits; \
+ \
+ u32 sub_tile_height; \
+ u32 column_data; \
+ \
+ u32 texture_mask = (psx_gpu->texture_mask_width & 0xF) | \
+ ((psx_gpu->texture_mask_height & 0xF) << 4) | \
+ ((psx_gpu->texture_mask_width >> 4) << 8) | \
+ ((psx_gpu->texture_mask_height >> 4) << 12); \
+ u32 texture_offset = ((v & 0xF) << 4) | ((u & 0xF0) << 4) | \
+ ((v & 0xF0) << 8); \
+ u32 texture_offset_base = texture_offset; \
+ u32 control_mask; \
+ \
+ u16 *fb_ptr = psx_gpu->vram_out_ptr + (y * 1024) + (x - offset_u); \
+ u32 num_blocks = psx_gpu->num_blocks; \
+ block_struct *block = psx_gpu->blocks + num_blocks; \
+ \
+ u16 *texture_block_ptr; \
+ vec_8x8u texels; \
+ \
+ setup_sprite_tiled_initialize_##texture_mode##x4mode(); \
+ \
+ control_mask = tile_width == 1; \
+ control_mask |= (tile_height == 1) << 1; \
+ control_mask |= setup_sprite_comapre_left_block_mask##x4mode() << 2; \
+ control_mask |= setup_sprite_comapre_right_block_mask##x4mode() << 3; \
+ \
+ switch(control_mask) \
+ { \
+ default: \
+ case 0x0: \
+ setup_sprite_tile_column_width_multi(texture_mode, multi, full, full, \
+ x4mode); \
+ break; \
+ \
+ case 0x1: \
+ setup_sprite_tile_column_width_single(texture_mode, multi, full, none, \
+ x4mode); \
+ break; \
+ \
+ case 0x2: \
+ setup_sprite_tile_column_width_multi(texture_mode, single, full, full, \
+ x4mode); \
+ break; \
+ \
+ case 0x3: \
+ setup_sprite_tile_column_width_single(texture_mode, single, full, none, \
+ x4mode); \
+ break; \
+ \
+ case 0x4: \
+ setup_sprite_tile_column_width_multi(texture_mode, multi, half, full, \
+ x4mode); \
+ break; \
+ \
+ case 0x5: \
+ setup_sprite_tile_column_width_single(texture_mode, multi, half, right, \
+ x4mode); \
+ break; \
+ \
+ case 0x6: \
+ setup_sprite_tile_column_width_multi(texture_mode, single, half, full, \
+ x4mode); \
+ break; \
+ \
+ case 0x7: \
+ setup_sprite_tile_column_width_single(texture_mode, single, half, right, \
+ x4mode); \
+ break; \
+ \
+ case 0x8: \
+ setup_sprite_tile_column_width_multi(texture_mode, multi, full, half, \
+ x4mode); \
+ break; \
+ \
+ case 0x9: \
+ setup_sprite_tile_column_width_single(texture_mode, multi, half, left, \
+ x4mode); \
+ break; \
+ \
+ case 0xA: \
+ setup_sprite_tile_column_width_multi(texture_mode, single, full, half, \
+ x4mode); \
+ break; \
+ \
+ case 0xB: \
+ setup_sprite_tile_column_width_single(texture_mode, single, half, left, \
+ x4mode); \
+ break; \
+ \
+ case 0xC: \
+ setup_sprite_tile_column_width_multi(texture_mode, multi, half, half, \
+ x4mode); \
+ break; \
+ \
+ case 0xE: \
+ setup_sprite_tile_column_width_multi(texture_mode, single, half, half, \
+ x4mode); \
+ break; \
+ } \
+
+void setup_sprite_4bpp(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
+ s32 width, s32 height, u32 color)
+{
+#if 0
+ setup_sprite_4bpp_(psx_gpu, x, y, u, v, width, height, color);
+ return;
+#endif
+ setup_sprite_tiled_do(4bpp,)
+}
+
+void setup_sprite_8bpp(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
+ s32 width, s32 height, u32 color)
+{
+#if 0
+ setup_sprite_8bpp_(psx_gpu, x, y, u, v, width, height, color);
+ return;
+#endif
+ setup_sprite_tiled_do(8bpp,)
+}
+
+#undef draw_mask_fb_ptr_left
+#undef draw_mask_fb_ptr_right
+
+void setup_sprite_4bpp_4x(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
+ s32 width, s32 height, u32 color)
+{
+#if 0
+ setup_sprite_4bpp_4x_(psx_gpu, x, y, u, v, width, height, color);
+ return;
+#endif
+ setup_sprite_tiled_do(4bpp, _4x)
+}
+
+void setup_sprite_8bpp_4x(psx_gpu_struct *psx_gpu, s32 x, s32 y, s32 u, s32 v,
+ s32 width, s32 height, u32 color)
+{
+#if 0
+ setup_sprite_8bpp_4x_(psx_gpu, x, y, u, v, width, height, color);
+ return;
+#endif
+ setup_sprite_tiled_do(8bpp, _4x)
+}
+
+
+void scale2x_tiles8(void * __restrict__ dst_, const void * __restrict__ src_, int w8, int h)
+{
+#if 0
+ scale2x_tiles8_(dst, src_, w8, h);
+ return;
+#endif
+ const u16 * __restrict__ src = src_;
+ const u16 * __restrict__ src1;
+ u16 * __restrict__ dst = dst_;
+ u16 * __restrict__ dst1;
+ gvreg a, b;
+ int w;
+ for (; h > 0; h--, src += 1024, dst += 1024*2)
+ {
+ src1 = src;
+ dst1 = dst;
+ for (w = w8; w > 0; w--, src1 += 8, dst1 += 8*2)
+ {
+ gvld1q_u16(a, src1);
+ gvzipq_u16(a, b, a, a);
+ gvst1q_u16(a, dst1);
+ gvst1q_u16(b, dst1 + 8);
+ gvst1q_u16(a, dst1 + 1024);
+ gvst1q_u16(b, dst1 + 1024 + 8);
+ }
+ }
+}
+
+// vim:ts=2:sw=2:expandtab