/**
* Parse TGSI shader and translate to SVGA/DX9 serialized
* representation.
*
* In this function SVGA shader is emitted to an in-memory buffer that
* can be dynamically grown. Once we've finished and know how large
* it is, it will be copied to a hardware buffer for upload.
*/
struct svga_shader_variant *
svga_tgsi_vgpu9_translate(struct svga_context *svga,
const struct svga_shader *shader,
const struct svga_compile_key *key,
enum pipe_shader_type unit)
{
struct svga_shader_variant *variant = NULL;
struct svga_shader_emitter emit;
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_TGSIVGPU9TRANSLATE);
memset(&emit, 0, sizeof(emit));
emit.size = 1024;
emit.buf = MALLOC(emit.size);
if (emit.buf == NULL) {
goto fail;
}
emit.ptr = emit.buf;
emit.unit = unit;
emit.key = *key;
tgsi_scan_shader(shader->tokens, &emit.info);
emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
if (unit == PIPE_SHADER_FRAGMENT)
emit.imm_start += key->num_unnormalized_coords;
if (unit == PIPE_SHADER_VERTEX) {
emit.imm_start += key->vs.need_prescale ? 2 : 0;
}
emit.nr_hw_float_const =
(emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
if (emit.nr_hw_temp >= SVGA3D_TEMPREG_MAX) {
debug_printf("svga: too many temporary registers (%u)\n",
emit.nr_hw_temp);
goto fail;
}
if (emit.info.indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
debug_printf(
"svga: indirect indexing of temporary registers is not supported.\n");
goto fail;
}
emit.in_main_func = TRUE;
if (!svga_shader_emit_header(&emit)) {
debug_printf("svga: emit header failed\n");
goto fail;
}
if (!svga_shader_emit_instructions(&emit, shader->tokens)) {
debug_printf("svga: emit instructions failed\n");
goto fail;
}
variant = svga_new_shader_variant(svga);
if (!variant)
goto fail;
variant->shader = shader;
variant->tokens = (const unsigned *) emit.buf;
variant->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
memcpy(&variant->key, key, sizeof(*key));
variant->id = UTIL_BITMASK_INVALID_INDEX;
variant->pstipple_sampler_unit = emit.pstipple_sampler_unit;
/* If there was exactly one write to a fragment shader output register
* and it came from a constant buffer, we know all fragments will have
* the same color (except for blending).
*/
variant->constant_color_output =
emit.constant_color_output && emit.num_output_writes == 1;
#if 0
if (!svga_shader_verify(variant->tokens, variant->nr_tokens) ||
SVGA_DEBUG & DEBUG_TGSI) {
debug_printf("#####################################\n");
debug_printf("Shader %u below\n", shader->id);
tgsi_dump(shader->tokens, 0);
if (SVGA_DEBUG & DEBUG_TGSI) {
debug_printf("Shader %u compiled below\n", shader->id);
svga_shader_dump(variant->tokens, variant->nr_tokens, FALSE);
}
debug_printf("#####################################\n");
}
#endif
goto done;
fail:
FREE(variant);
if (emit.buf != err_buf)
FREE(emit.buf);
variant = NULL;
done:
SVGA_STATS_TIME_POP(svga_sws(svga));
return variant;
}
/* Parse TGSI shader and translate to SVGA/DX9 serialized
* representation.
*
* In this function SVGA shader is emitted to an in-memory buffer that
* can be dynamically grown. Once we've finished and know how large
* it is, it will be copied to a hardware buffer for upload.
*/
static struct svga_shader_result *
svga_tgsi_translate( const struct svga_shader *shader,
union svga_compile_key key,
unsigned unit )
{
struct svga_shader_result *result = NULL;
struct svga_shader_emitter emit;
int ret = 0;
memset(&emit, 0, sizeof(emit));
emit.use_sm30 = shader->use_sm30;
emit.size = 1024;
emit.buf = MALLOC(emit.size);
if (emit.buf == NULL) {
ret = PIPE_ERROR_OUT_OF_MEMORY;
goto fail;
}
emit.ptr = emit.buf;
emit.unit = unit;
emit.key = key;
tgsi_scan_shader( shader->tokens, &emit.info);
emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
if (unit == PIPE_SHADER_FRAGMENT)
emit.imm_start += key.fkey.num_unnormalized_coords;
if (unit == PIPE_SHADER_VERTEX) {
emit.imm_start += key.vkey.need_prescale ? 2 : 0;
emit.imm_start += key.vkey.num_zero_stride_vertex_elements;
}
emit.nr_hw_const = (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
emit.in_main_func = TRUE;
if (!svga_shader_emit_header( &emit ))
goto fail;
if (!svga_shader_emit_instructions( &emit, shader->tokens ))
goto fail;
result = CALLOC_STRUCT(svga_shader_result);
if (result == NULL)
goto fail;
result->shader = shader;
result->tokens = (const unsigned *)emit.buf;
result->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
memcpy(&result->key, &key, sizeof key);
if (SVGA_DEBUG & DEBUG_TGSI)
{
debug_printf( "#####################################\n" );
debug_printf( "Shader %u below\n", shader->id );
tgsi_dump( shader->tokens, 0 );
if (SVGA_DEBUG & DEBUG_TGSI) {
debug_printf( "Shader %u compiled below\n", shader->id );
svga_shader_dump( result->tokens,
result->nr_tokens ,
FALSE );
}
debug_printf( "#####################################\n" );
}
return result;
fail:
FREE(result);
FREE(emit.buf);
return NULL;
}