radeonsi: add TGSI_SEMANTIC_CS_USER_DATA for reading up to 4 SGPRs with TGSI

diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index ea6fa3e..c5d3d5f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -128,6 +128,8 @@ static void si_create_compute_state_async(void *job, int thread_index)
        program->reads_variable_block_size =
                sel.info.uses_block_size &&
                sel.info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] == 0;
+       program->num_cs_user_data_dwords =
+               sel.info.properties[TGSI_PROPERTY_CS_USER_DATA_DWORDS];
 
        void *ir_binary = si_get_ir_binary(&sel);
 
@@ -159,7 +161,8 @@ static void si_create_compute_state_async(void *job, int thread_index)
                bool scratch_enabled = shader->config.scratch_bytes_per_wave > 0;
                unsigned user_sgprs = SI_NUM_RESOURCE_SGPRS +
                                      (sel.info.uses_grid_size ? 3 : 0) +
-                                     (program->reads_variable_block_size ? 3 : 0);
+                                     (program->reads_variable_block_size ? 3 : 0) +
+                                     program->num_cs_user_data_dwords;
 
                shader->config.rsrc1 =
                        S_00B848_VGPRS((shader->config.num_vgprs - 1) / 4) |
@@ -706,7 +709,7 @@ static bool si_upload_compute_input(struct si_context *sctx,
        return true;
 }
 
-static void si_setup_tgsi_grid(struct si_context *sctx,
+static void si_setup_tgsi_user_data(struct si_context *sctx,
                                 const struct pipe_grid_info *info)
 {
        struct si_compute *program = sctx->cs_shader_state.program;
@@ -716,6 +719,8 @@ static void si_setup_tgsi_grid(struct si_context *sctx,
        unsigned block_size_reg = grid_size_reg +
                                  /* 12 bytes = 3 dwords. */
                                  12 * program->uses_grid_size;
+       unsigned cs_user_data_reg = block_size_reg +
+                                   12 * program->reads_variable_block_size;
 
        if (info->indirect) {
                if (program->uses_grid_size) {
@@ -751,6 +756,11 @@ static void si_setup_tgsi_grid(struct si_context *sctx,
                        radeon_emit(cs, info->block[2]);
                }
        }
+
+       if (program->num_cs_user_data_dwords) {
+               radeon_set_sh_reg_seq(cs, cs_user_data_reg, program->num_cs_user_data_dwords);
+               radeon_emit_array(cs, sctx->cs_user_data, program->num_cs_user_data_dwords);
+       }
 }
 
 static void si_emit_dispatch_packets(struct si_context *sctx,
@@ -908,7 +918,7 @@ static void si_launch_grid(
        }
 
        if (program->ir_type != PIPE_SHADER_IR_NATIVE)
-               si_setup_tgsi_grid(sctx, info);
+               si_setup_tgsi_user_data(sctx, info);
 
        si_emit_dispatch_packets(sctx, info);
 

diff --git a/src/gallium/drivers/radeonsi/si_compute.h b/src/gallium/drivers/radeonsi/si_compute.h
index ef8b4ae..99b5016 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_compute.h
+++ b/src/gallium/drivers/radeonsi/si_compute.h
@@ -57,6 +57,7 @@ struct si_compute {
        unsigned uses_bindless_samplers:1;
        unsigned uses_bindless_images:1;
        bool reads_variable_block_size;
+       unsigned num_cs_user_data_dwords;
 };
 
 void si_destroy_compute(struct si_compute *program);

diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 4c3f13b..100d016 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -854,6 +854,7 @@ struct si_context {
        unsigned                        border_color_count;
        unsigned                        num_vs_blit_sgprs;
        uint32_t                        vs_blit_sh_data[SI_VS_BLIT_SGPRS_POS_TEXCOORD];
+       uint32_t                        cs_user_data[4];
 
        /* Vertex and index buffers. */
        bool                            vertex_buffers_dirty;

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index cfd99b6..c51e91b 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2267,6 +2267,10 @@ void si_load_system_value(struct si_shader_context *ctx,
                break;
        }
 
+       case TGSI_SEMANTIC_CS_USER_DATA:
+               value = LLVMGetParam(ctx->main_fn, ctx->param_cs_user_data);
+               break;
+
        default:
                assert(!"unknown system value");
                return;
@@ -4948,6 +4952,13 @@ static void create_function(struct si_shader_context *ctx)
                    shader->selector->info.properties[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH] == 0)
                        ctx->param_block_size = add_arg(&fninfo, ARG_SGPR, v3i32);
 
+               unsigned cs_user_data_dwords =
+                       shader->selector->info.properties[TGSI_PROPERTY_CS_USER_DATA_DWORDS];
+               if (cs_user_data_dwords) {
+                       ctx->param_cs_user_data = add_arg(&fninfo, ARG_SGPR,
+                                                         LLVMVectorType(ctx->i32, cs_user_data_dwords));
+               }
+
                for (i = 0; i < 3; i++) {
                        ctx->abi.workgroup_ids[i] = NULL;
                        if (shader->selector->info.uses_block_id[i])

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 2dc4bc7..14230b8 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -276,14 +276,21 @@ enum {
 
 /* SI-specific system values. */
 enum {
+       /* Values from set_tess_state. */
        TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI = TGSI_SEMANTIC_COUNT,
        TGSI_SEMANTIC_DEFAULT_TESSINNER_SI,
+
+       /* Up to 4 dwords in user SGPRs for compute shaders. */
+       TGSI_SEMANTIC_CS_USER_DATA,
 };
 
 enum {
        /* Use a property enum that CS wouldn't use. */
        TGSI_PROPERTY_CS_LOCAL_SIZE = TGSI_PROPERTY_FS_COORD_ORIGIN,
 
+       /* The number of used user data dwords in the range [1, 4]. */
+       TGSI_PROPERTY_CS_USER_DATA_DWORDS = TGSI_PROPERTY_FS_COORD_PIXEL_CENTER,
+
        /* Use a property enum that VS wouldn't use. */
        TGSI_PROPERTY_VS_BLIT_SGPRS = TGSI_PROPERTY_FS_COORD_ORIGIN,
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index aff7e04..a638dbf 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -173,6 +173,7 @@ struct si_shader_context {
        int param_gs_vtx45_offset; /* in dwords (GFX9) */
        /* CS */
        int param_block_size;
+       int param_cs_user_data;
 
        struct ac_llvm_compiler *compiler;