KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

radeonsi: get rid of secondary input/output word 

By keeping track of fewer generics, everything can fit into 64 bits.

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This commit is contained in:
Nicolai Hähnle 2017-05-02 17:46:12 +02:00
parent 0dd8aa44b3
commit a16ae77185
3 changed files with 13 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/gallium/drivers/radeonsi/si_shader.c
View File

@ -148,29 +148,22 @@ unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index)
assert(!"invalid generic index");
return 0;
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;
return SI_MAX_IO_GENERIC + 4;
case TGSI_SEMANTIC_LAYER:
return 1;
return SI_MAX_IO_GENERIC + 5;
case TGSI_SEMANTIC_VIEWPORT_INDEX:
return 2;
return SI_MAX_IO_GENERIC + 6;
case TGSI_SEMANTIC_PRIMID:
return 3;
return SI_MAX_IO_GENERIC + 7;
case TGSI_SEMANTIC_COLOR: /* these alias */
case TGSI_SEMANTIC_BCOLOR:
return 4 + index;
assert(index < 2);
return SI_MAX_IO_GENERIC + 8 + index;
case TGSI_SEMANTIC_TEXCOORD:
assert(index < 8);
return 6 + index;
assert(SI_MAX_IO_GENERIC + 10 + index < 64);
return SI_MAX_IO_GENERIC + 10 + index;
default:
assert(!"invalid semantic name");
return 0;
@ -2298,16 +2291,10 @@ static void si_llvm_export_vs(struct lp_build_tgsi_context *bld_base,
if (semantic_index >= SI_MAX_IO_GENERIC)
break;
/* fall through */
case TGSI_SEMANTIC_CLIPDIST:
default:
if (shader->key.opt.hw_vs.kill_outputs &
(1ull << si_shader_io_get_unique_index(semantic_name, semantic_index)))
export_param = false;
break;
default:
if (shader->key.opt.hw_vs.kill_outputs2 &
(1u << si_shader_io_get_unique_index2(semantic_name, semantic_index)))
export_param = false;
break;
}
if (outputs[i].vertex_stream[0] != 0 &&
@ -7159,7 +7146,6 @@ static void si_dump_shader_key(unsigned processor, const struct si_shader *shade
processor == PIPE_SHADER_VERTEX) &&
!key->as_es && !key->as_ls) {
fprintf(f, " opt.hw_vs.kill_outputs = 0x%"PRIx64"\n", key->opt.hw_vs.kill_outputs);
fprintf(f, " opt.hw_vs.kill_outputs2 = 0x%x\n", key->opt.hw_vs.kill_outputs2);
fprintf(f, " opt.hw_vs.clip_disable = %u\n", key->opt.hw_vs.clip_disable);
}
}

6
src/gallium/drivers/radeonsi/si_shader.h
View File

@ -361,11 +361,9 @@ struct si_shader_selector {
unsigned local_size;
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 */
uint32_t patch_outputs_written; /* "get_unique_index_patch" bits */
uint64_t inputs_read; /* "get_unique_index" bits */
uint32_t inputs_read2; /* "get_unique_index2" bits */
};
/* Valid shader configurations:
@ -505,7 +503,6 @@ struct si_shader_key {
struct {
struct {
uint64_t kill_outputs; /* "get_unique_index" bits */
uint32_t kill_outputs2; /* "get_unique_index2" bits */
unsigned clip_disable:1;
} hw_vs; /* HW VS (it can be VS, TES, GS) */
@ -604,7 +601,6 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
void si_shader_destroy(struct si_shader *shader);
unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name, unsigned index);
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, const struct si_shader *shader,
struct pipe_debug_callback *debug, unsigned processor,

22
src/gallium/drivers/radeonsi/si_state_shaders.c
View File

@ -1231,22 +1231,17 @@ static void si_shader_selector_key_hw_vs(struct si_context *sctx,
/* Find out which VS outputs aren't used by the PS. */
uint64_t outputs_written = vs->outputs_written;
uint32_t outputs_written2 = vs->outputs_written2;
uint64_t inputs_read = 0;
uint32_t inputs_read2 = 0;
outputs_written &= ~0x3; /* ignore POSITION, PSIZE */
if (!ps_disabled) {
inputs_read = ps->inputs_read;
inputs_read2 = ps->inputs_read2;
}
uint64_t linked = outputs_written & inputs_read;
uint32_t linked2 = outputs_written2 & inputs_read2;
key->opt.hw_vs.kill_outputs = ~linked & outputs_written;
key->opt.hw_vs.kill_outputs2 = ~linked2 & outputs_written2;
}
/* Compute the key for the hw shader variant */
@ -1846,7 +1841,7 @@ void si_init_shader_selector_async(void *job, int thread_index)
if (index >= SI_MAX_IO_GENERIC)
break;
/* fall through */
case TGSI_SEMANTIC_CLIPDIST:
default:
id = si_shader_io_get_unique_index(name, index);
sel->outputs_written &= ~(1ull << id);
break;
@ -1855,9 +1850,6 @@ void si_init_shader_selector_async(void *job, int thread_index)
case TGSI_SEMANTIC_CLIPVERTEX:
case TGSI_SEMANTIC_EDGEFLAG:
break;
default:
id = si_shader_io_get_unique_index2(name, index);
sel->outputs_written2 &= ~(1u << id);
}
}
}
@ -2003,18 +1995,13 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
if (index >= SI_MAX_IO_GENERIC)
break;
/* fall through */
case TGSI_SEMANTIC_POSITION:
case TGSI_SEMANTIC_PSIZE:
case TGSI_SEMANTIC_CLIPDIST:
default:
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;
@ -2037,15 +2024,12 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
if (index >= SI_MAX_IO_GENERIC)
break;
/* fall through */
case TGSI_SEMANTIC_CLIPDIST:
default:
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);
}
}