From: Dave Airlie <airl...@redhat.com>
This removes the other geometry specific user sgpr.
Signed-off-by: Dave Airlie <airl...@redhat.com>
---
src/amd/common/ac_nir_to_llvm.c | 37 +++++++++++++++++++------------------
src/amd/common/ac_nir_to_llvm.h | 1 -
src/amd/vulkan/radv_pipeline.c | 9 ---------
3 files changed, 19 insertions(+), 28 deletions(-)
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 9e4069b535a..43e13ec91e4 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -88,7 +88,6 @@ struct radv_shader_context {
LLVMValueRef tes_u;
LLVMValueRef tes_v;
- LLVMValueRef gsvs_ring_stride;
LLVMValueRef gs2vs_offset;
LLVMValueRef gs_wave_id;
LLVMValueRef gs_vtx_offset[6];
@@ -122,6 +121,8 @@ struct radv_shader_context {
uint32_t tcs_vertices_per_patch;
uint32_t tcs_num_inputs;
uint32_t tcs_num_patches;
+ uint32_t max_gsvs_emit_size;
+ uint32_t gsvs_vertex_size;
};
static inline struct radv_shader_context *
@@ -636,7 +637,6 @@ static void allocate_user_sgprs(struct radv_shader_context
*ctx,
user_sgpr_info->sgpr_count +=
count_vs_user_sgprs(ctx);
}
}
- user_sgpr_info->sgpr_count += 1;
break;
default:
break;
@@ -966,8 +966,6 @@ static void create_function(struct radv_shader_context *ctx,
&args);
}
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_ring_stride);
if (needs_view_index)
add_arg(&args, ARG_SGPR, ctx->ac.i32,
&ctx->abi.view_index);
@@ -995,8 +993,6 @@ static void create_function(struct radv_shader_context *ctx,
&user_sgpr_info, &args,
&desc_sets);
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_ring_stride);
if (needs_view_index)
add_arg(&args, ARG_SGPR, ctx->ac.i32,
&ctx->abi.view_index);
@@ -1122,8 +1118,6 @@ static void create_function(struct radv_shader_context
*ctx,
previous_stage,
&user_sgpr_idx);
}
- set_loc_shader(ctx, AC_UD_GS_VS_RING_STRIDE_ENTRIES,
- &user_sgpr_idx, 1);
if (ctx->abi.view_index)
set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx,
1);
break;
@@ -6740,6 +6734,8 @@ ac_setup_rings(struct radv_shader_context *ctx)
if (ctx->stage == MESA_SHADER_GEOMETRY) {
LLVMValueRef tmp;
uint32_t num_entries = 64;
+ LLVMValueRef gsvs_ring_stride = LLVMConstInt(ctx->ac.i32,
ctx->max_gsvs_emit_size, false);
+ LLVMValueRef gsvs_ring_desc = LLVMConstInt(ctx->ac.i32,
ctx->max_gsvs_emit_size << 16, false);
ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac,
ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_ESGS_GS, false));
ctx->gsvs_ring = ac_build_load_to_sgpr(&ctx->ac,
ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_GSVS_GS, false));
@@ -6747,10 +6743,10 @@ ac_setup_rings(struct radv_shader_context *ctx)
tmp = LLVMConstInt(ctx->ac.i32, num_entries, false);
if (ctx->options->chip_class >= VI)
- tmp = LLVMBuildMul(ctx->ac.builder,
LLVMBuildLShr(ctx->ac.builder, ctx->gsvs_ring_stride, LLVMConstInt(ctx->ac.i32,
16, false), ""), tmp, "");
+ tmp = LLVMBuildMul(ctx->ac.builder, gsvs_ring_stride,
tmp, "");
ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder,
ctx->gsvs_ring, tmp, LLVMConstInt(ctx->ac.i32, 2, false), "");
tmp = LLVMBuildExtractElement(ctx->ac.builder, ctx->gsvs_ring,
ctx->ac.i32_1, "");
- tmp = LLVMBuildOr(ctx->ac.builder, tmp, ctx->gsvs_ring_stride,
"");
+ tmp = LLVMBuildOr(ctx->ac.builder, tmp, gsvs_ring_desc, "");
ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder,
ctx->gsvs_ring, tmp, ctx->ac.i32_1, "");
}
@@ -6968,6 +6964,17 @@ LLVMModuleRef
ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
if (i)
emit_barrier(&ctx.ac, ctx.stage);
+ nir_foreach_variable(variable, &shaders[i]->outputs)
+ scan_shader_output_decl(&ctx, variable, shaders[i],
shaders[i]->info.stage);
+
+ if (shaders[i]->info.stage == MESA_SHADER_GEOMETRY) {
+ unsigned addclip =
shaders[i]->info.clip_distance_array_size +
+
shaders[i]->info.cull_distance_array_size > 4;
+ ctx.gsvs_vertex_size =
(util_bitcount64(ctx.output_mask) + addclip) * 16;
+ ctx.max_gsvs_emit_size = ctx.gsvs_vertex_size *
+ shaders[i]->info.gs.vertices_out;
+ }
+
ac_setup_rings(&ctx);
LLVMBasicBlockRef merge_block;
@@ -6994,9 +7001,6 @@ LLVMModuleRef
ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
else if(shader_count >= 2 && shaders[i]->info.stage ==
MESA_SHADER_GEOMETRY)
prepare_gs_input_vgprs(&ctx);
- nir_foreach_variable(variable, &shaders[i]->outputs)
- scan_shader_output_decl(&ctx, variable, shaders[i],
shaders[i]->info.stage);
-
ac_nir_translate(&ctx.ac, &ctx.abi, shaders[i]);
if (shader_count >= 2) {
@@ -7005,11 +7009,8 @@ LLVMModuleRef
ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
}
if (shaders[i]->info.stage == MESA_SHADER_GEOMETRY) {
- unsigned addclip =
shaders[i]->info.clip_distance_array_size +
-
shaders[i]->info.cull_distance_array_size > 4;
- shader_info->gs.gsvs_vertex_size =
(util_bitcount64(ctx.output_mask) + addclip) * 16;
- shader_info->gs.max_gsvs_emit_size =
shader_info->gs.gsvs_vertex_size *
- shaders[i]->info.gs.vertices_out;
+ shader_info->gs.gsvs_vertex_size = ctx.gsvs_vertex_size;
+ shader_info->gs.max_gsvs_emit_size =
ctx.max_gsvs_emit_size;
} else if (shaders[i]->info.stage == MESA_SHADER_TESS_CTRL) {
shader_info->tcs.num_patches = ctx.tcs_num_patches;
shader_info->tcs.lds_size =
calculate_tess_lds_size(&ctx);
diff --git a/src/amd/common/ac_nir_to_llvm.h b/src/amd/common/ac_nir_to_llvm.h
index f4825ef4cd5..0ea6832acdc 100644
--- a/src/amd/common/ac_nir_to_llvm.h
+++ b/src/amd/common/ac_nir_to_llvm.h
@@ -112,7 +112,6 @@ enum ac_ud_index {
AC_UD_PS_MAX_UD,
AC_UD_CS_GRID_SIZE = AC_UD_SHADER_START,
AC_UD_CS_MAX_UD,
- AC_UD_GS_VS_RING_STRIDE_ENTRIES = AC_UD_VS_MAX_UD,
AC_UD_GS_MAX_UD,
AC_UD_TCS_MAX_UD,
AC_UD_TES_MAX_UD,
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index 5d1b5f6e352..6c247589348 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -2617,15 +2617,6 @@ radv_pipeline_generate_geometry_shader(struct
radeon_winsys_cs *cs,
}
radv_pipeline_generate_hw_vs(cs, pipeline, pipeline->gs_copy_shader);
-
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline,
MESA_SHADER_GEOMETRY,
-
AC_UD_GS_VS_RING_STRIDE_ENTRIES);
- if (loc->sgpr_idx != -1) {
- uint32_t stride = gs->info.gs.max_gsvs_emit_size;
- stride = S_008F04_STRIDE(stride);
- radeon_set_sh_reg_seq(cs, R_00B230_SPI_SHADER_USER_DATA_GS_0 +
loc->sgpr_idx * 4, 1);
- radeon_emit(cs, stride);
- }
}
static uint32_t offset_to_ps_input(uint32_t offset, bool flat_shade)
--
2.14.3
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
