radeonsi/compute: Upload work group, work item size in input buffer
authorTom Stellard <thomas.stellard@amd.com>
Tue, 14 May 2013 02:13:53 +0000 (22:13 -0400)
committerTom Stellard <thomas.stellard@amd.com>
Mon, 3 Jun 2013 18:03:13 +0000 (14:03 -0400)
src/gallium/drivers/radeonsi/radeonsi_compute.c

index 4341ecc0924e204684d0aa67a70f80d6f801d8db..ed5158752ca336038e7693860fba3010e89613b8 100644 (file)
@@ -91,9 +91,12 @@ static void radeonsi_launch_grid(
        struct r600_context *rctx = (struct r600_context*)ctx;
        struct si_pipe_compute *program = rctx->cs_shader_state.program;
        struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
-       struct si_resource *input_buffer = NULL;
-       uint32_t input_offset = 0;
-       uint64_t input_va;
+       struct si_resource *kernel_args_buffer = NULL;
+       unsigned kernel_args_size;
+       unsigned num_work_size_bytes = 36;
+       uint32_t kernel_args_offset = 0;
+       uint32_t *kernel_args;
+       uint64_t kernel_args_va;
        uint64_t shader_va;
        unsigned arg_user_sgpr_count = 2;
        unsigned i;
@@ -112,16 +115,29 @@ static void radeonsi_launch_grid(
        si_pm4_inval_shader_cache(pm4);
        si_cmd_surface_sync(pm4, pm4->cp_coher_cntl);
 
-       /* Upload the input data */
-       r600_upload_const_buffer(rctx, &input_buffer, input,
-                                       program->input_size, &input_offset);
-       input_va = r600_resource_va(ctx->screen, (struct pipe_resource*)input_buffer);
-       input_va += input_offset;
+       /* Upload the kernel arguments */
 
-       si_pm4_add_bo(pm4, input_buffer, RADEON_USAGE_READ);
+       /* The extra num_work_size_bytes are for work group / work item size information */
+       kernel_args_size = program->input_size + num_work_size_bytes;
+       kernel_args = MALLOC(kernel_args_size);
+       for (i = 0; i < 3; i++) {
+               kernel_args[i] = grid_layout[i];
+               kernel_args[i + 3] = grid_layout[i] * block_layout[i];
+               kernel_args[i + 6] = block_layout[i];
+       }
+
+       memcpy(kernel_args + (num_work_size_bytes / 4), input, program->input_size);
+
+       r600_upload_const_buffer(rctx, &kernel_args_buffer, (uint8_t*)kernel_args,
+                                       kernel_args_size, &kernel_args_offset);
+       kernel_args_va = r600_resource_va(ctx->screen,
+                               (struct pipe_resource*)kernel_args_buffer);
+       kernel_args_va += kernel_args_offset;
+
+       si_pm4_add_bo(pm4, kernel_args_buffer, RADEON_USAGE_READ);
 
-       si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0, input_va);
-       si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0 + 4, S_008F04_BASE_ADDRESS_HI (input_va >> 32) | S_008F04_STRIDE(0));
+       si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0, kernel_args_va);
+       si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0 + 4, S_008F04_BASE_ADDRESS_HI (kernel_args_va >> 32) | S_008F04_STRIDE(0));
 
        si_pm4_set_reg(pm4, R_00B810_COMPUTE_START_X, 0);
        si_pm4_set_reg(pm4, R_00B814_COMPUTE_START_Y, 0);