[Mesa-dev] [PATCH 04/16] radeonsi/gfx9: move RW_BUFFERS from s[0:1] to s[8:9] for HS and GS
Marek Olšák
maraeo at gmail.com
Fri Oct 13 12:04:00 UTC 2017
From: Marek Olšák <marek.olsak at amd.com>
Let's use the same user data SGPRs in all stages.
(for SPI_SHADER_USER_DATA_COMMON_0)
---
src/gallium/drivers/radeonsi/si_descriptors.c | 24 ++++-----------------
src/gallium/drivers/radeonsi/si_shader.c | 31 +++++++++++----------------
2 files changed, 16 insertions(+), 39 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index cda7d94..9ba8df6 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -1973,47 +1973,31 @@ static void si_emit_shader_pointer(struct si_context *sctx,
radeon_emit(cs, va >> 32);
}
static void si_emit_global_shader_pointers(struct si_context *sctx,
struct si_descriptors *descs)
{
si_emit_shader_pointer(sctx, descs,
R_00B030_SPI_SHADER_USER_DATA_PS_0);
si_emit_shader_pointer(sctx, descs,
R_00B130_SPI_SHADER_USER_DATA_VS_0);
+ si_emit_shader_pointer(sctx, descs,
+ R_00B330_SPI_SHADER_USER_DATA_ES_0);
if (sctx->b.chip_class >= GFX9) {
/* GFX9 merged LS-HS and ES-GS. */
- if (descs == &sctx->descriptors[SI_DESCS_RW_BUFFERS]) {
- /* Set RW_BUFFERS in the special registers, so that
- * it's preloaded into s[0:1] instead of s[8:9].
- */
- si_emit_shader_pointer(sctx, descs,
- R_00B208_SPI_SHADER_USER_DATA_ADDR_LO_GS);
- si_emit_shader_pointer(sctx, descs,
- R_00B408_SPI_SHADER_USER_DATA_ADDR_LO_HS);
- } else {
- /* Set BINDLESS_SAMPLERS_AND_IMAGES into s[10:11],
- * s[8:9] remains unused for now.
- */
- assert(descs == &sctx->bindless_descriptors);
- si_emit_shader_pointer(sctx, descs,
- R_00B330_SPI_SHADER_USER_DATA_ES_0);
- si_emit_shader_pointer(sctx, descs,
- R_00B430_SPI_SHADER_USER_DATA_LS_0);
- }
+ si_emit_shader_pointer(sctx, descs,
+ R_00B430_SPI_SHADER_USER_DATA_LS_0);
} else {
si_emit_shader_pointer(sctx, descs,
R_00B230_SPI_SHADER_USER_DATA_GS_0);
si_emit_shader_pointer(sctx, descs,
- R_00B330_SPI_SHADER_USER_DATA_ES_0);
- si_emit_shader_pointer(sctx, descs,
R_00B430_SPI_SHADER_USER_DATA_HS_0);
si_emit_shader_pointer(sctx, descs,
R_00B530_SPI_SHADER_USER_DATA_LS_0);
}
}
void si_emit_graphics_shader_pointers(struct si_context *sctx,
struct r600_atom *atom)
{
unsigned mask;
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 4456c3d..62a056d 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -3024,25 +3024,27 @@ static void si_llvm_emit_tcs_epilogue(struct lp_build_tgsi_context *bld_base)
ret = LLVMBuildInsertValue(builder, ret, tf_lds_offset, vgpr++, "");
}
ctx->return_value = ret;
}
/* Pass TCS inputs from LS to TCS on GFX9. */
static void si_set_ls_return_value_for_tcs(struct si_shader_context *ctx)
{
LLVMValueRef ret = ctx->return_value;
- ret = si_insert_input_ptr_as_2xi32(ctx, ret, ctx->param_rw_buffers, 0);
ret = si_insert_input_ret(ctx, ret, ctx->param_tcs_offchip_offset, 2);
ret = si_insert_input_ret(ctx, ret, ctx->param_merged_wave_info, 3);
ret = si_insert_input_ret(ctx, ret, ctx->param_tcs_factor_offset, 4);
ret = si_insert_input_ret(ctx, ret, ctx->param_merged_scratch_offset, 5);
+
+ ret = si_insert_input_ptr_as_2xi32(ctx, ret, ctx->param_rw_buffers,
+ 8 + SI_SGPR_RW_BUFFERS);
ret = si_insert_input_ptr_as_2xi32(ctx, ret,
ctx->param_bindless_samplers_and_images,
8 + SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES);
ret = si_insert_input_ret(ctx, ret, ctx->param_vs_state_bits,
8 + SI_SGPR_VS_STATE_BITS);
ret = si_insert_input_ret(ctx, ret, ctx->param_tcs_offchip_layout,
8 + GFX9_SGPR_TCS_OFFCHIP_LAYOUT);
ret = si_insert_input_ret(ctx, ret, ctx->param_tcs_out_lds_offsets,
8 + GFX9_SGPR_TCS_OUT_OFFSETS);
@@ -3065,25 +3067,26 @@ static void si_set_ls_return_value_for_tcs(struct si_shader_context *ctx)
ret = si_insert_input_ret_float(ctx, ret,
ctx->param_tcs_rel_ids, vgpr++);
ctx->return_value = ret;
}
/* Pass GS inputs from ES to GS on GFX9. */
static void si_set_es_return_value_for_gs(struct si_shader_context *ctx)
{
LLVMValueRef ret = ctx->return_value;
- ret = si_insert_input_ptr_as_2xi32(ctx, ret, ctx->param_rw_buffers, 0);
ret = si_insert_input_ret(ctx, ret, ctx->param_gs2vs_offset, 2);
ret = si_insert_input_ret(ctx, ret, ctx->param_merged_wave_info, 3);
-
ret = si_insert_input_ret(ctx, ret, ctx->param_merged_scratch_offset, 5);
+
+ ret = si_insert_input_ptr_as_2xi32(ctx, ret, ctx->param_rw_buffers,
+ 8 + SI_SGPR_RW_BUFFERS);
ret = si_insert_input_ptr_as_2xi32(ctx, ret,
ctx->param_bindless_samplers_and_images,
8 + SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES);
unsigned desc_param = ctx->param_vs_state_bits + 1;
ret = si_insert_input_ptr_as_2xi32(ctx, ret, desc_param,
8 + GFX9_SGPR_GS_CONST_AND_SHADER_BUFFERS);
ret = si_insert_input_ptr_as_2xi32(ctx, ret, desc_param + 1,
8 + GFX9_SGPR_GS_SAMPLERS_AND_IMAGES);
@@ -4430,35 +4433,30 @@ static void create_function(struct si_shader_context *ctx)
* placed after the user SGPRs.
*/
for (i = 0; i < GFX6_TCS_NUM_USER_SGPR + 2; i++)
returns[num_returns++] = ctx->i32; /* SGPRs */
for (i = 0; i < 11; i++)
returns[num_returns++] = ctx->f32; /* VGPRs */
break;
case SI_SHADER_MERGED_VERTEX_TESSCTRL:
/* Merged stages have 8 system SGPRs at the beginning. */
- ctx->param_rw_buffers = /* SPI_SHADER_USER_DATA_ADDR_LO_HS */
- add_arg(&fninfo, ARG_SGPR, si_const_array(ctx->v4i32, SI_NUM_RW_BUFFERS));
+ add_arg(&fninfo, ARG_SGPR, ctx->i32); /* SPI_SHADER_USER_DATA_ADDR_LO_HS */
+ add_arg(&fninfo, ARG_SGPR, ctx->i32); /* SPI_SHADER_USER_DATA_ADDR_HI_HS */
ctx->param_tcs_offchip_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_merged_wave_info = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_factor_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_merged_scratch_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
- add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
- add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
-
- ctx->param_bindless_samplers_and_images =
- add_arg(&fninfo, ARG_SGPR, si_const_array(ctx->v8i32, 0));
-
+ declare_global_desc_pointers(ctx, &fninfo);
declare_per_stage_desc_pointers(ctx, &fninfo,
ctx->type == PIPE_SHADER_VERTEX);
declare_vs_specific_input_sgprs(ctx, &fninfo);
ctx->param_tcs_offchip_layout = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_out_lds_offsets = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_out_lds_layout = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_offchip_addr_base64k = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_factor_addr_base64k = add_arg(&fninfo, ARG_SGPR, ctx->i32);
add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
@@ -4488,35 +4486,30 @@ static void create_function(struct si_shader_context *ctx)
*/
for (i = 0; i <= 8 + GFX9_SGPR_TCS_FACTOR_ADDR_BASE64K; i++)
returns[num_returns++] = ctx->i32; /* SGPRs */
for (i = 0; i < 11; i++)
returns[num_returns++] = ctx->f32; /* VGPRs */
}
break;
case SI_SHADER_MERGED_VERTEX_OR_TESSEVAL_GEOMETRY:
/* Merged stages have 8 system SGPRs at the beginning. */
- ctx->param_rw_buffers = /* SPI_SHADER_USER_DATA_ADDR_LO_GS */
- add_arg(&fninfo, ARG_SGPR, si_const_array(ctx->v4i32, SI_NUM_RW_BUFFERS));
+ add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused (SPI_SHADER_USER_DATA_ADDR_LO_GS) */
+ add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused (SPI_SHADER_USER_DATA_ADDR_HI_GS) */
ctx->param_gs2vs_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_merged_wave_info = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_offchip_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_merged_scratch_offset = add_arg(&fninfo, ARG_SGPR, ctx->i32);
add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused (SPI_SHADER_PGM_LO/HI_GS << 8) */
add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused (SPI_SHADER_PGM_LO/HI_GS >> 24) */
- add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
- add_arg(&fninfo, ARG_SGPR, ctx->i32); /* unused */
-
- ctx->param_bindless_samplers_and_images =
- add_arg(&fninfo, ARG_SGPR, si_const_array(ctx->v8i32, 0));
-
+ declare_global_desc_pointers(ctx, &fninfo);
declare_per_stage_desc_pointers(ctx, &fninfo,
(ctx->type == PIPE_SHADER_VERTEX ||
ctx->type == PIPE_SHADER_TESS_EVAL));
if (ctx->type == PIPE_SHADER_VERTEX) {
declare_vs_specific_input_sgprs(ctx, &fninfo);
} else {
/* TESS_EVAL (and also GEOMETRY):
* Declare as many input SGPRs as the VS has. */
ctx->param_tcs_offchip_layout = add_arg(&fninfo, ARG_SGPR, ctx->i32);
ctx->param_tcs_offchip_addr_base64k = add_arg(&fninfo, ARG_SGPR, ctx->i32);
--
2.7.4
More information about the mesa-dev
mailing list