1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 **********************************************************/
27 #include "pipe/p_compiler.h"
28 #include "pipe/p_shader_tokens.h"
29 #include "pipe/p_defines.h"
30 #include "tgsi/tgsi_parse.h"
31 #include "tgsi/tgsi_dump.h"
32 #include "tgsi/tgsi_scan.h"
33 #include "util/u_math.h"
34 #include "util/u_memory.h"
35 #include "util/u_bitmask.h"
37 #include "svgadump/svga_shader_dump.h"
39 #include "svga_context.h"
40 #include "svga_tgsi.h"
41 #include "svga_tgsi_emit.h"
42 #include "svga_debug.h"
44 #include "svga_hw_reg.h"
45 #include "svga3d_shaderdefs.h"
48 /* Sinkhole used only in error conditions.
50 static char err_buf[128];
53 static void svga_destroy_shader_emitter( struct svga_shader_emitter *emit )
55 if (emit->buf != err_buf)
61 static boolean svga_shader_expand( struct svga_shader_emitter *emit )
64 unsigned newsize = emit->size * 2;
66 if(emit->buf != err_buf)
67 new_buf = REALLOC(emit->buf, emit->size, newsize);
71 if (new_buf == NULL) {
74 emit->size = sizeof(err_buf);
79 emit->ptr = new_buf + (emit->ptr - emit->buf);
84 static INLINE boolean reserve( struct svga_shader_emitter *emit,
87 if (emit->ptr - emit->buf + nr_dwords * sizeof(unsigned) >= emit->size) {
88 if (!svga_shader_expand( emit ))
95 boolean svga_shader_emit_dword( struct svga_shader_emitter *emit,
98 if (!reserve(emit, 1))
101 *(unsigned *)emit->ptr = dword;
102 emit->ptr += sizeof dword;
106 boolean svga_shader_emit_dwords( struct svga_shader_emitter *emit,
107 const unsigned *dwords,
110 if (!reserve(emit, nr))
113 memcpy( emit->ptr, dwords, nr * sizeof *dwords );
114 emit->ptr += nr * sizeof *dwords;
118 boolean svga_shader_emit_opcode( struct svga_shader_emitter *emit,
121 SVGA3dShaderInstToken *here;
123 if (!reserve(emit, 1))
126 here = (SVGA3dShaderInstToken *)emit->ptr;
127 here->value = opcode;
129 if (emit->insn_offset) {
130 SVGA3dShaderInstToken *prev = (SVGA3dShaderInstToken *)(emit->buf +
132 prev->size = (here - prev) - 1;
135 emit->insn_offset = emit->ptr - emit->buf;
136 emit->ptr += sizeof(unsigned);
141 static boolean svga_shader_emit_header( struct svga_shader_emitter *emit )
143 SVGA3dShaderVersion header;
145 memset( &header, 0, sizeof header );
147 switch (emit->unit) {
148 case PIPE_SHADER_FRAGMENT:
149 header.value = SVGA3D_PS_30;
151 case PIPE_SHADER_VERTEX:
152 header.value = SVGA3D_VS_30;
156 return svga_shader_emit_dword( emit, header.value );
161 * Use the shader info to generate a bitmask indicating which generic
162 * inputs are used by the shader. A set bit indicates that GENERIC[i]
166 svga_get_generic_inputs_mask(const struct tgsi_shader_info *info)
168 unsigned i, mask = 0x0;
170 for (i = 0; i < info->num_inputs; i++) {
171 if (info->input_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
172 unsigned j = info->input_semantic_index[i];
173 assert(j < sizeof(mask) * 8);
183 * Given a mask of used generic variables (as returned by the above functions)
184 * fill in a table which maps those indexes to small integers.
185 * This table is used by the remap_generic_index() function in
186 * svga_tgsi_decl_sm30.c
187 * Example: if generics_mask = binary(1010) it means that GENERIC[1] and
188 * GENERIC[3] are used. The remap_table will contain:
191 * The remaining table entries will be filled in with the next unused
192 * generic index (in this example, 2).
195 svga_remap_generics(unsigned generics_mask,
196 int8_t remap_table[MAX_GENERIC_VARYING])
198 /* Note texcoord[0] is reserved so start at 1 */
199 unsigned count = 1, i;
201 for (i = 0; i < MAX_GENERIC_VARYING; i++) {
205 /* for each bit set in generic_mask */
206 while (generics_mask) {
207 unsigned index = ffs(generics_mask) - 1;
208 remap_table[index] = count++;
209 generics_mask &= ~(1 << index);
215 * Use the generic remap table to map a TGSI generic varying variable
216 * index to a small integer. If the remapping table doesn't have a
217 * valid value for the given index (the table entry is -1) it means
218 * the fragment shader doesn't use that VS output. Just allocate
219 * the next free value in that case. Alternately, we could cull
220 * VS instructions that write to register, or replace the register
221 * with a dummy temp register.
222 * XXX TODO: we should do one of the later as it would save precious
223 * texcoord registers.
226 svga_remap_generic_index(int8_t remap_table[MAX_GENERIC_VARYING],
229 assert(generic_index < MAX_GENERIC_VARYING);
231 if (generic_index >= MAX_GENERIC_VARYING) {
232 /* just don't return a random/garbage value */
233 generic_index = MAX_GENERIC_VARYING - 1;
236 if (remap_table[generic_index] == -1) {
237 /* This is a VS output that has no matching PS input. Find a
241 for (i = 0; i < MAX_GENERIC_VARYING; i++) {
242 max = MAX2(max, remap_table[i]);
244 remap_table[generic_index] = max + 1;
247 return remap_table[generic_index];
251 /* Parse TGSI shader and translate to SVGA/DX9 serialized
254 * In this function SVGA shader is emitted to an in-memory buffer that
255 * can be dynamically grown. Once we've finished and know how large
256 * it is, it will be copied to a hardware buffer for upload.
258 static struct svga_shader_result *
259 svga_tgsi_translate( const struct svga_shader *shader,
260 struct svga_compile_key key,
263 struct svga_shader_result *result = NULL;
264 struct svga_shader_emitter emit;
266 memset(&emit, 0, sizeof(emit));
269 emit.buf = MALLOC(emit.size);
270 if (emit.buf == NULL) {
278 tgsi_scan_shader( shader->tokens, &emit.info);
280 emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
282 if (unit == PIPE_SHADER_FRAGMENT)
283 emit.imm_start += key.fkey.num_unnormalized_coords;
285 if (unit == PIPE_SHADER_VERTEX) {
286 emit.imm_start += key.vkey.need_prescale ? 2 : 0;
287 emit.imm_start += key.vkey.num_zero_stride_vertex_elements;
290 emit.nr_hw_float_const = (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
292 emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
293 emit.in_main_func = TRUE;
295 if (!svga_shader_emit_header( &emit ))
298 if (!svga_shader_emit_instructions( &emit, shader->tokens ))
301 result = CALLOC_STRUCT(svga_shader_result);
305 result->shader = shader;
306 result->tokens = (const unsigned *)emit.buf;
307 result->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
308 memcpy(&result->key, &key, sizeof key);
309 result->id = UTIL_BITMASK_INVALID_INDEX;
311 if (SVGA_DEBUG & DEBUG_TGSI)
313 debug_printf( "#####################################\n" );
314 debug_printf( "Shader %u below\n", shader->id );
315 tgsi_dump( shader->tokens, 0 );
316 if (SVGA_DEBUG & DEBUG_TGSI) {
317 debug_printf( "Shader %u compiled below\n", shader->id );
318 svga_shader_dump( result->tokens,
322 debug_printf( "#####################################\n" );
336 struct svga_shader_result *
337 svga_translate_fragment_program( const struct svga_fragment_shader *fs,
338 const struct svga_fs_compile_key *fkey )
340 struct svga_compile_key key;
342 memcpy(&key.fkey, fkey, sizeof *fkey);
344 memcpy(key.generic_remap_table, fs->generic_remap_table,
345 sizeof(fs->generic_remap_table));
347 return svga_tgsi_translate( &fs->base,
349 PIPE_SHADER_FRAGMENT );
352 struct svga_shader_result *
353 svga_translate_vertex_program( const struct svga_vertex_shader *vs,
354 const struct svga_vs_compile_key *vkey )
356 struct svga_compile_key key;
358 memcpy(&key.vkey, vkey, sizeof *vkey);
360 /* Note: we could alternately store the remap table in the vkey but
361 * that would make it larger. We just regenerate it here instead.
363 svga_remap_generics(vkey->fs_generic_inputs, key.generic_remap_table);
365 return svga_tgsi_translate( &vs->base,
367 PIPE_SHADER_VERTEX );
371 void svga_destroy_shader_result( struct svga_shader_result *result )
373 FREE((unsigned *)result->tokens);