14a49301eSmrg/**********************************************************
24a49301eSmrg * Copyright 2008-2009 VMware, Inc.  All rights reserved.
34a49301eSmrg *
44a49301eSmrg * Permission is hereby granted, free of charge, to any person
54a49301eSmrg * obtaining a copy of this software and associated documentation
64a49301eSmrg * files (the "Software"), to deal in the Software without
74a49301eSmrg * restriction, including without limitation the rights to use, copy,
84a49301eSmrg * modify, merge, publish, distribute, sublicense, and/or sell copies
94a49301eSmrg * of the Software, and to permit persons to whom the Software is
104a49301eSmrg * furnished to do so, subject to the following conditions:
114a49301eSmrg *
124a49301eSmrg * The above copyright notice and this permission notice shall be
134a49301eSmrg * included in all copies or substantial portions of the Software.
144a49301eSmrg *
154a49301eSmrg * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
164a49301eSmrg * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
174a49301eSmrg * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
184a49301eSmrg * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
194a49301eSmrg * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
204a49301eSmrg * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
214a49301eSmrg * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
224a49301eSmrg * SOFTWARE.
234a49301eSmrg *
244a49301eSmrg **********************************************************/
254a49301eSmrg
264a49301eSmrg
274a49301eSmrg#include "pipe/p_compiler.h"
284a49301eSmrg#include "pipe/p_shader_tokens.h"
294a49301eSmrg#include "pipe/p_defines.h"
304a49301eSmrg#include "tgsi/tgsi_parse.h"
314a49301eSmrg#include "tgsi/tgsi_dump.h"
324a49301eSmrg#include "tgsi/tgsi_scan.h"
33af69d88dSmrg#include "util/u_math.h"
344a49301eSmrg#include "util/u_memory.h"
354a49301eSmrg#include "util/u_bitmask.h"
364a49301eSmrg
374a49301eSmrg#include "svgadump/svga_shader_dump.h"
384a49301eSmrg
394a49301eSmrg#include "svga_context.h"
4001e04c3fSmrg#include "svga_shader.h"
414a49301eSmrg#include "svga_tgsi.h"
424a49301eSmrg#include "svga_tgsi_emit.h"
434a49301eSmrg#include "svga_debug.h"
444a49301eSmrg
454a49301eSmrg#include "svga_hw_reg.h"
464a49301eSmrg#include "svga3d_shaderdefs.h"
474a49301eSmrg
484a49301eSmrg
494a49301eSmrg/* Sinkhole used only in error conditions.
504a49301eSmrg */
514a49301eSmrgstatic char err_buf[128];
524a49301eSmrg
534a49301eSmrg
54af69d88dSmrgstatic boolean
55af69d88dSmrgsvga_shader_expand(struct svga_shader_emitter *emit)
564a49301eSmrg{
574a49301eSmrg   char *new_buf;
584a49301eSmrg   unsigned newsize = emit->size * 2;
594a49301eSmrg
60af69d88dSmrg   if (emit->buf != err_buf)
614a49301eSmrg      new_buf = REALLOC(emit->buf, emit->size, newsize);
624a49301eSmrg   else
634a49301eSmrg      new_buf = NULL;
644a49301eSmrg
6501e04c3fSmrg   if (!new_buf) {
664a49301eSmrg      emit->ptr = err_buf;
674a49301eSmrg      emit->buf = err_buf;
684a49301eSmrg      emit->size = sizeof(err_buf);
694a49301eSmrg      return FALSE;
704a49301eSmrg   }
714a49301eSmrg
724a49301eSmrg   emit->size = newsize;
734a49301eSmrg   emit->ptr = new_buf + (emit->ptr - emit->buf);
744a49301eSmrg   emit->buf = new_buf;
754a49301eSmrg   return TRUE;
76af69d88dSmrg}
774a49301eSmrg
78af69d88dSmrg
7901e04c3fSmrgstatic inline boolean
80af69d88dSmrgreserve(struct svga_shader_emitter *emit, unsigned nr_dwords)
814a49301eSmrg{
824a49301eSmrg   if (emit->ptr - emit->buf + nr_dwords * sizeof(unsigned) >= emit->size) {
83af69d88dSmrg      if (!svga_shader_expand(emit)) {
844a49301eSmrg         return FALSE;
85af69d88dSmrg      }
864a49301eSmrg   }
874a49301eSmrg
884a49301eSmrg   return TRUE;
894a49301eSmrg}
904a49301eSmrg
91af69d88dSmrg
92af69d88dSmrgboolean
93af69d88dSmrgsvga_shader_emit_dword(struct svga_shader_emitter * emit, unsigned dword)
944a49301eSmrg{
954a49301eSmrg   if (!reserve(emit, 1))
964a49301eSmrg      return FALSE;
974a49301eSmrg
98af69d88dSmrg   *(unsigned *) emit->ptr = dword;
994a49301eSmrg   emit->ptr += sizeof dword;
1004a49301eSmrg   return TRUE;
1014a49301eSmrg}
1024a49301eSmrg
103af69d88dSmrg
104af69d88dSmrgboolean
105af69d88dSmrgsvga_shader_emit_dwords(struct svga_shader_emitter * emit,
106af69d88dSmrg                        const unsigned *dwords, unsigned nr)
1074a49301eSmrg{
1084a49301eSmrg   if (!reserve(emit, nr))
1094a49301eSmrg      return FALSE;
1104a49301eSmrg
111af69d88dSmrg   memcpy(emit->ptr, dwords, nr * sizeof *dwords);
1124a49301eSmrg   emit->ptr += nr * sizeof *dwords;
1134a49301eSmrg   return TRUE;
1144a49301eSmrg}
1154a49301eSmrg
116af69d88dSmrg
117af69d88dSmrgboolean
118af69d88dSmrgsvga_shader_emit_opcode(struct svga_shader_emitter * emit, unsigned opcode)
1194a49301eSmrg{
1204a49301eSmrg   SVGA3dShaderInstToken *here;
1214a49301eSmrg
1224a49301eSmrg   if (!reserve(emit, 1))
1234a49301eSmrg      return FALSE;
1244a49301eSmrg
125af69d88dSmrg   here = (SVGA3dShaderInstToken *) emit->ptr;
1264a49301eSmrg   here->value = opcode;
1274a49301eSmrg
1284a49301eSmrg   if (emit->insn_offset) {
129af69d88dSmrg      SVGA3dShaderInstToken *prev =
130af69d88dSmrg         (SVGA3dShaderInstToken *) (emit->buf + emit->insn_offset);
1314a49301eSmrg      prev->size = (here - prev) - 1;
1324a49301eSmrg   }
133af69d88dSmrg
1344a49301eSmrg   emit->insn_offset = emit->ptr - emit->buf;
1354a49301eSmrg   emit->ptr += sizeof(unsigned);
1364a49301eSmrg   return TRUE;
1374a49301eSmrg}
1384a49301eSmrg
1394a49301eSmrg
140af69d88dSmrgstatic boolean
141af69d88dSmrgsvga_shader_emit_header(struct svga_shader_emitter *emit)
1424a49301eSmrg{
1434a49301eSmrg   SVGA3dShaderVersion header;
1444a49301eSmrg
145af69d88dSmrg   memset(&header, 0, sizeof header);
1464a49301eSmrg
1474a49301eSmrg   switch (emit->unit) {
1484a49301eSmrg   case PIPE_SHADER_FRAGMENT:
149af69d88dSmrg      header.value = SVGA3D_PS_30;
1504a49301eSmrg      break;
1514a49301eSmrg   case PIPE_SHADER_VERTEX:
152af69d88dSmrg      header.value = SVGA3D_VS_30;
1534a49301eSmrg      break;
1544a49301eSmrg   }
155af69d88dSmrg
156af69d88dSmrg   return svga_shader_emit_dword(emit, header.value);
157af69d88dSmrg}
158af69d88dSmrg
159af69d88dSmrg
160af69d88dSmrg/**
161af69d88dSmrg * Parse TGSI shader and translate to SVGA/DX9 serialized
162af69d88dSmrg * representation.
1634a49301eSmrg *
1644a49301eSmrg * In this function SVGA shader is emitted to an in-memory buffer that
1654a49301eSmrg * can be dynamically grown.  Once we've finished and know how large
1664a49301eSmrg * it is, it will be copied to a hardware buffer for upload.
1674a49301eSmrg */
16801e04c3fSmrgstruct svga_shader_variant *
16901e04c3fSmrgsvga_tgsi_vgpu9_translate(struct svga_context *svga,
17001e04c3fSmrg                          const struct svga_shader *shader,
17101e04c3fSmrg                          const struct svga_compile_key *key,
17201e04c3fSmrg                          enum pipe_shader_type unit)
1734a49301eSmrg{
174af69d88dSmrg   struct svga_shader_variant *variant = NULL;
1754a49301eSmrg   struct svga_shader_emitter emit;
1764a49301eSmrg
17701e04c3fSmrg   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_TGSIVGPU9TRANSLATE);
17801e04c3fSmrg
1794a49301eSmrg   memset(&emit, 0, sizeof(emit));
1804a49301eSmrg
1814a49301eSmrg   emit.size = 1024;
1824a49301eSmrg   emit.buf = MALLOC(emit.size);
1834a49301eSmrg   if (emit.buf == NULL) {
1844a49301eSmrg      goto fail;
1854a49301eSmrg   }
1864a49301eSmrg
1874a49301eSmrg   emit.ptr = emit.buf;
1884a49301eSmrg   emit.unit = unit;
189af69d88dSmrg   emit.key = *key;
1904a49301eSmrg
191af69d88dSmrg   tgsi_scan_shader(shader->tokens, &emit.info);
1924a49301eSmrg
1934a49301eSmrg   emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
194af69d88dSmrg
1954a49301eSmrg   if (unit == PIPE_SHADER_FRAGMENT)
19601e04c3fSmrg      emit.imm_start += key->num_unnormalized_coords;
1974a49301eSmrg
1984a49301eSmrg   if (unit == PIPE_SHADER_VERTEX) {
19901e04c3fSmrg      emit.imm_start += key->vs.need_prescale ? 2 : 0;
2004a49301eSmrg   }
2014a49301eSmrg
202af69d88dSmrg   emit.nr_hw_float_const =
203af69d88dSmrg      (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
2044a49301eSmrg
2054a49301eSmrg   emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
206af69d88dSmrg
207af69d88dSmrg   if (emit.nr_hw_temp >= SVGA3D_TEMPREG_MAX) {
208af69d88dSmrg      debug_printf("svga: too many temporary registers (%u)\n",
209af69d88dSmrg                   emit.nr_hw_temp);
210af69d88dSmrg      goto fail;
211af69d88dSmrg   }
212af69d88dSmrg
21301e04c3fSmrg   if (emit.info.indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
21401e04c3fSmrg      debug_printf(
21501e04c3fSmrg         "svga: indirect indexing of temporary registers is not supported.\n");
21601e04c3fSmrg      goto fail;
21701e04c3fSmrg   }
21801e04c3fSmrg
2194a49301eSmrg   emit.in_main_func = TRUE;
2204a49301eSmrg
221af69d88dSmrg   if (!svga_shader_emit_header(&emit)) {
222af69d88dSmrg      debug_printf("svga: emit header failed\n");
2234a49301eSmrg      goto fail;
224af69d88dSmrg   }
2254a49301eSmrg
226af69d88dSmrg   if (!svga_shader_emit_instructions(&emit, shader->tokens)) {
227af69d88dSmrg      debug_printf("svga: emit instructions failed\n");
2284a49301eSmrg      goto fail;
229af69d88dSmrg   }
230af69d88dSmrg
2319f464c52Smaya   variant = svga_new_shader_variant(svga, unit);
23201e04c3fSmrg   if (!variant)
2334a49301eSmrg      goto fail;
2344a49301eSmrg
235af69d88dSmrg   variant->shader = shader;
236af69d88dSmrg   variant->tokens = (const unsigned *) emit.buf;
237af69d88dSmrg   variant->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
238af69d88dSmrg   memcpy(&variant->key, key, sizeof(*key));
239af69d88dSmrg   variant->id = UTIL_BITMASK_INVALID_INDEX;
240af69d88dSmrg
2417ec681f3Smrg   if (unit == PIPE_SHADER_FRAGMENT) {
2427ec681f3Smrg      struct svga_fs_variant *fs_variant = svga_fs_variant(variant);
2437ec681f3Smrg
2447ec681f3Smrg      fs_variant->pstipple_sampler_unit = emit.pstipple_sampler_unit;
2457ec681f3Smrg
2467ec681f3Smrg      /* If there was exactly one write to a fragment shader output register
2477ec681f3Smrg       * and it came from a constant buffer, we know all fragments will have
2487ec681f3Smrg       * the same color (except for blending).
2497ec681f3Smrg       */
2507ec681f3Smrg      fs_variant->constant_color_output =
2517ec681f3Smrg         emit.constant_color_output && emit.num_output_writes == 1;
2527ec681f3Smrg   }
25301e04c3fSmrg
25401e04c3fSmrg#if 0
25501e04c3fSmrg   if (!svga_shader_verify(variant->tokens, variant->nr_tokens) ||
25601e04c3fSmrg       SVGA_DEBUG & DEBUG_TGSI) {
257af69d88dSmrg      debug_printf("#####################################\n");
258af69d88dSmrg      debug_printf("Shader %u below\n", shader->id);
259af69d88dSmrg      tgsi_dump(shader->tokens, 0);
2604a49301eSmrg      if (SVGA_DEBUG & DEBUG_TGSI) {
261af69d88dSmrg         debug_printf("Shader %u compiled below\n", shader->id);
262af69d88dSmrg         svga_shader_dump(variant->tokens, variant->nr_tokens, FALSE);
2634a49301eSmrg      }
264af69d88dSmrg      debug_printf("#####################################\n");
2654a49301eSmrg   }
26601e04c3fSmrg#endif
2674a49301eSmrg
26801e04c3fSmrg   goto done;
2694a49301eSmrg
27001e04c3fSmrgfail:
271af69d88dSmrg   FREE(variant);
27201e04c3fSmrg   if (emit.buf != err_buf)
27301e04c3fSmrg      FREE(emit.buf);
27401e04c3fSmrg   variant = NULL;
2754a49301eSmrg
27601e04c3fSmrgdone:
27701e04c3fSmrg   SVGA_STATS_TIME_POP(svga_sws(svga));
27801e04c3fSmrg   return variant;
2794a49301eSmrg}
280