From: Marek Olšák <marek.ol...@amd.com>
It's just tgsi_shader_info with DEFAULT_VAL varyings removed.
---
src/gallium/drivers/radeonsi/si_shader.c | 22 ++++++++
src/gallium/drivers/radeonsi/si_shader.h | 10 ++--
src/gallium/drivers/radeonsi/si_state_shaders.c | 69 +++++++++++++++++++++++++
3 files changed, 98 insertions(+), 3 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index b250053..f8de049 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -134,20 +134,42 @@ unsigned si_shader_io_get_unique_index(unsigned
semantic_name, unsigned index)
return 1;
case TGSI_SEMANTIC_PATCH:
return 2 + index;
default:
assert(!"invalid semantic name");
return 0;
}
}
+unsigned si_shader_io_get_unique_index2(unsigned name, unsigned index)
+{
+ switch (name) {
+ case TGSI_SEMANTIC_FOG:
+ return 0;
+ case TGSI_SEMANTIC_LAYER:
+ return 1;
+ case TGSI_SEMANTIC_VIEWPORT_INDEX:
+ return 2;
+ case TGSI_SEMANTIC_PRIMID:
+ return 3;
+ case TGSI_SEMANTIC_COLOR: /* these alias */
+ case TGSI_SEMANTIC_BCOLOR:
+ return 4 + index;
+ case TGSI_SEMANTIC_TEXCOORD:
+ return 6 + index;
+ default:
+ assert(!"invalid semantic name");
+ return 0;
+ }
+}
+
/**
* Get the value of a shader input parameter and extract a bitfield.
*/
static LLVMValueRef unpack_param(struct si_shader_context *ctx,
unsigned param, unsigned rshift,
unsigned bitwidth)
{
struct gallivm_state *gallivm = &ctx->gallivm;
LLVMValueRef value = LLVMGetParam(ctx->main_fn,
param);
diff --git a/src/gallium/drivers/radeonsi/si_shader.h
b/src/gallium/drivers/radeonsi/si_shader.h
index 4cbd1c2..fc9c913 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -288,23 +288,26 @@ struct si_shader_selector {
unsigned color_attr_index[2];
unsigned db_shader_control;
/* Set 0xf or 0x0 (4 bits) per each written output.
* ANDed with spi_shader_col_format.
*/
unsigned colors_written_4bit;
/* CS parameters */
unsigned local_size;
- /* masks of "get_unique_index" bits */
- uint64_t outputs_written;
- uint32_t patch_outputs_written;
+ uint64_t outputs_written; /* "get_unique_index" bits */
+ uint32_t patch_outputs_written; /* "get_unique_index" bits */
+ uint32_t outputs_written2; /* "get_unique_index2" bits */
+
+ uint64_t inputs_read; /* "get_unique_index" bits */
+ uint32_t inputs_read2; /* "get_unique_index2" bits */
};
/* Valid shader configurations:
*
* API shaders VS | TCS | TES | GS |pass| PS
* are compiled as: | | | |thru|
* | | | | |
* Only VS & PS: VS | -- | -- | -- | -- | PS
* With GS: ES | -- | -- | GS | VS | PS
* With Tessel.: LS | HS | VS | -- | -- | PS
@@ -532,20 +535,21 @@ int si_shader_create(struct si_screen *sscreen,
LLVMTargetMachineRef tm,
int si_compile_llvm(struct si_screen *sscreen,
struct radeon_shader_binary *binary,
struct si_shader_config *conf,
LLVMTargetMachineRef tm,
LLVMModuleRef mod,
struct pipe_debug_callback *debug,
unsigned processor,
const char *name);
void si_shader_destroy(struct si_shader *shader);
unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index);
+unsigned si_shader_io_get_unique_index2(unsigned name, unsigned index);
int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader
*shader);
void si_shader_dump(struct si_screen *sscreen, struct si_shader *shader,
struct pipe_debug_callback *debug, unsigned processor,
FILE *f);
void si_shader_apply_scratch_relocs(struct si_context *sctx,
struct si_shader *shader,
struct si_shader_config *config,
uint64_t scratch_va);
void si_shader_binary_read_config(struct radeon_shader_binary *binary,
struct si_shader_config *conf,
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c
b/src/gallium/drivers/radeonsi/si_state_shaders.c
index d10084d..e4d8747 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1289,20 +1289,65 @@ void si_init_shader_selector_async(void *job, int
thread_index)
if (tgsi_binary) {
pipe_mutex_lock(sscreen->shader_cache_mutex);
if (!si_shader_cache_insert_shader(sscreen,
tgsi_binary, shader))
FREE(tgsi_binary);
pipe_mutex_unlock(sscreen->shader_cache_mutex);
}
}
sel->main_shader_part = shader;
+
+ /* Unset "outputs_written" flags for outputs converted to
+ * DEFAULT_VAL, so that later inter-shader optimizations don't
+ * try to eliminate outputs that don't exist in the final
+ * shader.
+ *
+ * This is only done if non-monolithic shaders are enabled.
+ */
+ if ((sel->type == PIPE_SHADER_VERTEX ||
+ sel->type == PIPE_SHADER_TESS_EVAL) &&
+ !shader->key.as_ls &&
+ !shader->key.as_es) {
+ unsigned i;
+
+ for (i = 0; i < sel->info.num_outputs; i++) {
+ unsigned offset =
shader->info.vs_output_param_offset[i];
+
+ if (offset <= EXP_PARAM_OFFSET_31)
+ continue;
+
+ unsigned name =
sel->info.output_semantic_name[i];
+ unsigned index =
sel->info.output_semantic_index[i];
+ unsigned id;
+
+ switch (name) {
+ case TGSI_SEMANTIC_GENERIC:
+ /* don't process indices the function
can't handle */
+ if (index >= 60)
+ break;
+ /* fall through */
+ case TGSI_SEMANTIC_CLIPDIST:
+ id =
si_shader_io_get_unique_index(name, index);
+ sel->outputs_written &= ~(1ull << id);
+ break;
+ case TGSI_SEMANTIC_POSITION: /* ignore these */
+ case TGSI_SEMANTIC_PSIZE:
+ case TGSI_SEMANTIC_CLIPVERTEX:
+ case TGSI_SEMANTIC_EDGEFLAG:
+ break;
+ default:
+ id =
si_shader_io_get_unique_index2(name, index);
+ sel->outputs_written2 &= ~(1u << id);
+ }
+ }
+ }
}
/* Pre-compilation. */
if (sscreen->b.debug_flags & DBG_PRECOMPILE) {
struct si_shader_ctx_state state = {sel};
struct si_shader_key key;
memset(&key, 0, sizeof(key));
si_parse_next_shader_property(&sel->info, &key);
@@ -1430,26 +1475,50 @@ static void *si_create_shader_selector(struct
pipe_context *ctx,
/* don't process indices the function can't
handle */
if (index >= 60)
break;
/* fall through */
case TGSI_SEMANTIC_POSITION:
case TGSI_SEMANTIC_PSIZE:
case TGSI_SEMANTIC_CLIPDIST:
sel->outputs_written |=
1llu <<
si_shader_io_get_unique_index(name, index);
break;
+ case TGSI_SEMANTIC_CLIPVERTEX: /* ignore these */
+ case TGSI_SEMANTIC_EDGEFLAG:
+ break;
+ default:
+ sel->outputs_written2 |=
+ 1u <<
si_shader_io_get_unique_index2(name, index);
}
}
sel->esgs_itemsize = util_last_bit64(sel->outputs_written) * 16;
break;
case PIPE_SHADER_FRAGMENT:
+ for (i = 0; i < sel->info.num_inputs; i++) {
+ unsigned name = sel->info.input_semantic_name[i];
+ unsigned index = sel->info.input_semantic_index[i];
+
+ switch (name) {
+ case TGSI_SEMANTIC_CLIPDIST:
+ case TGSI_SEMANTIC_GENERIC:
+ sel->inputs_read |=
+ 1llu <<
si_shader_io_get_unique_index(name, index);
+ break;
+ case TGSI_SEMANTIC_PCOORD: /* ignore this */
+ break;
+ default:
+ sel->inputs_read2 |=
+ 1u <<
si_shader_io_get_unique_index2(name, index);
+ }
+ }
+
for (i = 0; i < 8; i++)
if (sel->info.colors_written & (1 << i))
sel->colors_written_4bit |= 0xf << (4 * i);
for (i = 0; i < sel->info.num_inputs; i++) {
if (sel->info.input_semantic_name[i] ==
TGSI_SEMANTIC_COLOR) {
int index = sel->info.input_semantic_index[i];
sel->color_attr_index[index] = i;
}
}
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
