[Mesa-dev] [PATCH 1/2] radeonsi: split per-patch from per-vertex indices
Marek Olšák
maraeo at gmail.com
Fri May 5 17:36:10 UTC 2017
Reviewed-by: Marek Olšák <marek.olsak at amd.com>
Marek
On Wed, May 3, 2017 at 3:54 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>
> Make it a bit clearer that the index spaces are logically seperate by
> having them defined in different functions.
> ---
> src/gallium/drivers/radeonsi/si_shader.c | 56 +++++++++++++++++--------
> src/gallium/drivers/radeonsi/si_shader.h | 1 +
> src/gallium/drivers/radeonsi/si_state_shaders.c | 6 +--
> 3 files changed, 42 insertions(+), 21 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
> index 77dd6b1..a48a552 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.c
> +++ b/src/gallium/drivers/radeonsi/si_shader.c
> @@ -98,20 +98,42 @@ static bool is_merged_shader(struct si_shader *shader)
> if (shader->selector->screen->b.chip_class <= VI)
> return false;
>
> return shader->key.as_ls ||
> shader->key.as_es ||
> shader->selector->type == PIPE_SHADER_TESS_CTRL ||
> shader->selector->type == PIPE_SHADER_GEOMETRY;
> }
>
> /**
> + * Returns a unique index for a per-patch semantic name and index. The index
> + * must be less than 32, so that a 32-bit bitmask of used inputs or outputs
> + * can be calculated.
> + */
> +unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name, unsigned index)
> +{
> + switch (semantic_name) {
> + case TGSI_SEMANTIC_TESSOUTER:
> + return 0;
> + case TGSI_SEMANTIC_TESSINNER:
> + return 1;
> + case TGSI_SEMANTIC_PATCH:
> + assert(index < 30);
> + return 2 + index;
> +
> + default:
> + assert(!"invalid semantic name");
> + return 0;
> + }
> +}
> +
> +/**
> * Returns a unique index for a semantic name and index. The index must be
> * less than 64, so that a 64-bit bitmask of used inputs or outputs can be
> * calculated.
> */
> unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index)
> {
> switch (semantic_name) {
> case TGSI_SEMANTIC_POSITION:
> return 0;
> case TGSI_SEMANTIC_PSIZE:
> @@ -119,28 +141,20 @@ unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index)
> case TGSI_SEMANTIC_CLIPDIST:
> assert(index <= 1);
> return 2 + index;
> case TGSI_SEMANTIC_GENERIC:
> if (index <= 63-4)
> return 4 + index;
>
> assert(!"invalid generic index");
> return 0;
>
> - /* patch indices are completely separate and thus start from 0 */
> - case TGSI_SEMANTIC_TESSOUTER:
> - return 0;
> - case TGSI_SEMANTIC_TESSINNER:
> - 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:
> @@ -670,24 +684,29 @@ static LLVMValueRef get_dw_address(struct si_shader_context *ctx,
> else
> first = reg.Register.Index;
>
> ind_index = get_indirect_index(ctx, ®.Indirect,
> reg.Register.Index - first);
>
> base_addr = LLVMBuildAdd(gallivm->builder, base_addr,
> LLVMBuildMul(gallivm->builder, ind_index,
> LLVMConstInt(ctx->i32, 4, 0), ""), "");
>
> - param = si_shader_io_get_unique_index(name[first], index[first]);
> + param = reg.Register.Dimension ?
> + si_shader_io_get_unique_index(name[first], index[first]) :
> + si_shader_io_get_unique_index_patch(name[first], index[first]);
> } else {
> - param = si_shader_io_get_unique_index(name[reg.Register.Index],
> - index[reg.Register.Index]);
> + param = reg.Register.Dimension ?
> + si_shader_io_get_unique_index(name[reg.Register.Index],
> + index[reg.Register.Index]) :
> + si_shader_io_get_unique_index_patch(name[reg.Register.Index],
> + index[reg.Register.Index]);
> }
>
> /* Add the base address of the element. */
> return LLVMBuildAdd(gallivm->builder, base_addr,
> LLVMConstInt(ctx->i32, param * 4, 0), "");
> }
>
> /* The offchip buffer layout for TCS->TES is
> *
> * - attribute 0 of patch 0 vertex 0
> @@ -795,22 +814,23 @@ static LLVMValueRef get_tcs_tes_buffer_address_from_reg(
> param_base = reg.Register.Index;
>
> param_index = get_indirect_index(ctx, ®.Indirect,
> reg.Register.Index - param_base);
>
> } else {
> param_base = reg.Register.Index;
> param_index = ctx->i32_0;
> }
>
> - param_index_base = si_shader_io_get_unique_index(name[param_base],
> - index[param_base]);
> + param_index_base = reg.Register.Dimension ?
> + si_shader_io_get_unique_index(name[param_base], index[param_base]) :
> + si_shader_io_get_unique_index_patch(name[param_base], index[param_base]);
>
> param_index = LLVMBuildAdd(gallivm->builder, param_index,
> LLVMConstInt(ctx->i32, param_index_base, 0),
> "");
>
> return get_tcs_tes_buffer_address(ctx, get_rel_patch_id(ctx),
> vertex_index, param_index);
> }
>
> static LLVMValueRef buffer_load(struct lp_build_tgsi_context *bld_base,
> @@ -1548,21 +1568,21 @@ static void declare_system_value(struct si_shader_context *ctx,
> else if (ctx->type == PIPE_SHADER_TESS_EVAL)
> value = unpack_param(ctx, ctx->param_tcs_offchip_layout, 6, 6);
> else
> assert(!"invalid shader stage for TGSI_SEMANTIC_VERTICESIN");
> break;
>
> case TGSI_SEMANTIC_TESSINNER:
> case TGSI_SEMANTIC_TESSOUTER:
> {
> LLVMValueRef buffer, base, addr;
> - int param = si_shader_io_get_unique_index(decl->Semantic.Name, 0);
> + int param = si_shader_io_get_unique_index_patch(decl->Semantic.Name, 0);
>
> buffer = desc_from_addr_base64k(ctx, ctx->param_tcs_offchip_addr_base64k);
>
> base = LLVMGetParam(ctx->main_fn, ctx->param_tcs_offchip_offset);
> addr = get_tcs_tes_buffer_address(ctx, get_rel_patch_id(ctx), NULL,
> LLVMConstInt(ctx->i32, param, 0));
>
> value = buffer_load(&ctx->bld_base, TGSI_TYPE_FLOAT,
> ~0, buffer, base, addr, true);
>
> @@ -2537,22 +2557,22 @@ static void si_write_tess_factors(struct lp_build_tgsi_context *bld_base,
> inner_comps = 2;
> break;
> default:
> assert(0);
> return;
> }
>
> /* Load tess_inner and tess_outer from LDS.
> * Any invocation can write them, so we can't get them from a temporary.
> */
> - tess_inner_index = si_shader_io_get_unique_index(TGSI_SEMANTIC_TESSINNER, 0);
> - tess_outer_index = si_shader_io_get_unique_index(TGSI_SEMANTIC_TESSOUTER, 0);
> + tess_inner_index = si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSINNER, 0);
> + tess_outer_index = si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSOUTER, 0);
>
> lds_base = tcs_out_current_patch_data_offset;
> lds_inner = LLVMBuildAdd(gallivm->builder, lds_base,
> LLVMConstInt(ctx->i32,
> tess_inner_index * 4, 0), "");
> lds_outer = LLVMBuildAdd(gallivm->builder, lds_base,
> LLVMConstInt(ctx->i32,
> tess_outer_index * 4, 0), "");
>
> for (i = 0; i < 4; i++) {
> @@ -2621,33 +2641,33 @@ static void si_write_tess_factors(struct lp_build_tgsi_context *bld_base,
>
> /* Store the tess factors into the offchip buffer if TES reads them. */
> if (shader->key.part.tcs.epilog.tes_reads_tess_factors) {
> LLVMValueRef buf, base, inner_vec, outer_vec, tf_outer_offset;
> LLVMValueRef tf_inner_offset;
> unsigned param_outer, param_inner;
>
> buf = desc_from_addr_base64k(ctx, ctx->param_tcs_offchip_addr_base64k);
> base = LLVMGetParam(ctx->main_fn, ctx->param_tcs_offchip_offset);
>
> - param_outer = si_shader_io_get_unique_index(
> + param_outer = si_shader_io_get_unique_index_patch(
> TGSI_SEMANTIC_TESSOUTER, 0);
> tf_outer_offset = get_tcs_tes_buffer_address(ctx, rel_patch_id, NULL,
> LLVMConstInt(ctx->i32, param_outer, 0));
>
> outer_vec = lp_build_gather_values(gallivm, outer,
> util_next_power_of_two(outer_comps));
>
> ac_build_buffer_store_dword(&ctx->ac, buf, outer_vec,
> outer_comps, tf_outer_offset,
> base, 0, 1, 0, true, false);
> if (inner_comps) {
> - param_inner = si_shader_io_get_unique_index(
> + param_inner = si_shader_io_get_unique_index_patch(
> TGSI_SEMANTIC_TESSINNER, 0);
> tf_inner_offset = get_tcs_tes_buffer_address(ctx, rel_patch_id, NULL,
> LLVMConstInt(ctx->i32, param_inner, 0));
>
> inner_vec = inner_comps == 1 ? inner[0] :
> lp_build_gather_values(gallivm, inner, inner_comps);
> ac_build_buffer_store_dword(&ctx->ac, buf, inner_vec,
> inner_comps, tf_inner_offset,
> base, 0, 1, 0, true, false);
> }
> diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
> index 0988d91..cb8a902 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.h
> +++ b/src/gallium/drivers/radeonsi/si_shader.h
> @@ -598,20 +598,21 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
> struct pipe_debug_callback *debug);
> int si_compile_llvm(struct si_screen *sscreen,
> struct ac_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_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, struct si_shader *shader,
> struct pipe_debug_callback *debug, unsigned processor,
> FILE *f, bool check_debug_option);
> void si_multiwave_lds_size_workaround(struct si_screen *sscreen,
> unsigned *lds_size);
> void si_shader_apply_scratch_relocs(struct si_context *sctx,
> struct si_shader *shader,
> diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
> index e479422..68f4d21 100644
> --- a/src/gallium/drivers/radeonsi/si_state_shaders.c
> +++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
> @@ -1975,35 +1975,35 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
> sel->max_gs_stream = MAX2(sel->max_gs_stream,
> sel->so.output[i].stream);
>
> sel->gs_input_verts_per_prim =
> u_vertices_per_prim(sel->info.properties[TGSI_PROPERTY_GS_INPUT_PRIM]);
> break;
>
> case PIPE_SHADER_TESS_CTRL:
> /* Always reserve space for these. */
> sel->patch_outputs_written |=
> - (1llu << si_shader_io_get_unique_index(TGSI_SEMANTIC_TESSINNER, 0)) |
> - (1llu << si_shader_io_get_unique_index(TGSI_SEMANTIC_TESSOUTER, 0));
> + (1llu << si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSINNER, 0)) |
> + (1llu << si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSOUTER, 0));
> /* fall through */
> case PIPE_SHADER_VERTEX:
> case PIPE_SHADER_TESS_EVAL:
> for (i = 0; i < sel->info.num_outputs; i++) {
> unsigned name = sel->info.output_semantic_name[i];
> unsigned index = sel->info.output_semantic_index[i];
>
> switch (name) {
> case TGSI_SEMANTIC_TESSINNER:
> case TGSI_SEMANTIC_TESSOUTER:
> case TGSI_SEMANTIC_PATCH:
> sel->patch_outputs_written |=
> - 1llu << si_shader_io_get_unique_index(name, index);
> + 1llu << si_shader_io_get_unique_index_patch(name, index);
> break;
>
> case TGSI_SEMANTIC_GENERIC:
> /* 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:
> --
> 2.9.3
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
More information about the mesa-dev
mailing list