[Mesa-dev] [PATCH 05/14] radv: drop tcs_out_layout

[Dave Airlie]

From: Dave Airlie <airlied at redhat.com>

Move all calculations to shader generation.

Signed-off-by: Dave Airlie <airlied at redhat.com>
---
 src/amd/vulkan/radv_nir_to_llvm.c | 22 +++++++++++++---------
 src/amd/vulkan/radv_pipeline.c    |  8 ++------
 2 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/src/amd/vulkan/radv_nir_to_llvm.c b/src/amd/vulkan/radv_nir_to_llvm.c
index ee32549bde6..ad165f17cf0 100644
--- a/src/amd/vulkan/radv_nir_to_llvm.c
+++ b/src/amd/vulkan/radv_nir_to_llvm.c
@@ -64,7 +64,6 @@ struct radv_shader_context {
 
 	LLVMValueRef tcs_offchip_layout;
 	LLVMValueRef tcs_out_offsets;
-	LLVMValueRef tcs_out_layout;
 	LLVMValueRef oc_lds;
 	LLVMValueRef merged_wave_info;
 	LLVMValueRef tess_factor_offset;
@@ -172,13 +171,22 @@ get_tcs_in_patch_stride(struct radv_shader_context *ctx)
 static LLVMValueRef
 get_tcs_out_patch_stride(struct radv_shader_context *ctx)
 {
-	return ac_unpack_param(&ctx->ac, ctx->tcs_out_layout, 0, 13);
+	uint32_t num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
+	uint32_t num_tcs_patch_outputs = util_last_bit64(ctx->shader_info->info.tcs.patch_outputs_written);
+	uint32_t output_vertex_size = num_tcs_outputs * 16;
+	uint32_t pervertex_output_patch_size = ctx->tcs_vertices_per_patch * output_vertex_size;
+	uint32_t output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16;
+	output_patch_size /= 4;
+	return LLVMConstInt(ctx->ac.i32, output_patch_size, false);
 }
 
 static LLVMValueRef
 get_tcs_out_vertex_stride(struct radv_shader_context *ctx)
 {
-	return ac_unpack_param(&ctx->ac, ctx->tcs_out_layout, 13, 8);
+	uint32_t num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
+	uint32_t output_vertex_size = num_tcs_outputs * 16;
+	output_vertex_size /= 4;
+	return LLVMConstInt(ctx->ac.i32, output_vertex_size, false);
 }
 
 static LLVMValueRef
@@ -466,7 +474,7 @@ static void allocate_user_sgprs(struct radv_shader_context *ctx,
 			if (previous_stage == MESA_SHADER_VERTEX)
 				user_sgpr_info->sgpr_count += count_vs_user_sgprs(ctx);
 		}
-		user_sgpr_info->sgpr_count += 3;
+		user_sgpr_info->sgpr_count += 2;
 		break;
 	case MESA_SHADER_TESS_EVAL:
 		user_sgpr_info->sgpr_count += 1;
@@ -766,8 +774,6 @@ static void create_function(struct radv_shader_context *ctx,
 				&ctx->tcs_offchip_layout);
 			add_arg(&args, ARG_SGPR, ctx->ac.i32,
 				&ctx->tcs_out_offsets);
-			add_arg(&args, ARG_SGPR, ctx->ac.i32,
-				&ctx->tcs_out_layout);
 			if (needs_view_index)
 				add_arg(&args, ARG_SGPR, ctx->ac.i32,
 					&ctx->abi.view_index);
@@ -789,8 +795,6 @@ static void create_function(struct radv_shader_context *ctx,
 				&ctx->tcs_offchip_layout);
 			add_arg(&args, ARG_SGPR, ctx->ac.i32,
 				&ctx->tcs_out_offsets);
-			add_arg(&args, ARG_SGPR, ctx->ac.i32,
-				&ctx->tcs_out_layout);
 			if (needs_view_index)
 				add_arg(&args, ARG_SGPR, ctx->ac.i32,
 					&ctx->abi.view_index);
@@ -999,7 +1003,7 @@ static void create_function(struct radv_shader_context *ctx,
 	case MESA_SHADER_TESS_CTRL:
 		set_vs_specific_input_locs(ctx, stage, has_previous_stage,
 					   previous_stage, &user_sgpr_idx);
-		set_loc_shader(ctx, AC_UD_TCS_OFFCHIP_LAYOUT, &user_sgpr_idx, 3);
+		set_loc_shader(ctx, AC_UD_TCS_OFFCHIP_LAYOUT, &user_sgpr_idx, 2);
 		if (ctx->abi.view_index)
 			set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
 		break;
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index 557d5ec58b3..01808cc3f55 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -62,7 +62,6 @@ struct radv_blend_state {
 
 struct radv_tessellation_state {
 	uint32_t ls_hs_config;
-	uint32_t tcs_out_layout;
 	uint32_t tcs_out_offsets;
 	uint32_t offchip_layout;
 	unsigned num_patches;
@@ -1382,8 +1381,6 @@ calculate_tess_state(struct radv_pipeline *pipeline,
 
 	tess.lds_size = lds_size;
 
-	tess.tcs_out_layout = (output_patch_size / 4) |
-		((output_vertex_size / 4) << 13);
 	tess.tcs_out_offsets = (output_patch0_offset / 16) |
 		((perpatch_output_offset / 16) << 16);
 	tess.offchip_layout = (pervertex_output_patch_size * num_patches << 16) |
@@ -2615,12 +2612,11 @@ radv_pipeline_generate_tess_shaders(struct radeon_winsys_cs *cs,
 	loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_TESS_CTRL, AC_UD_TCS_OFFCHIP_LAYOUT);
 	if (loc->sgpr_idx != -1) {
 		uint32_t base_reg = pipeline->user_data_0[MESA_SHADER_TESS_CTRL];
-		assert(loc->num_sgprs == 3);
+		assert(loc->num_sgprs == 2);
 		assert(!loc->indirect);
-		radeon_set_sh_reg_seq(cs, base_reg + loc->sgpr_idx * 4, 3);
+		radeon_set_sh_reg_seq(cs, base_reg + loc->sgpr_idx * 4, 2);
 		radeon_emit(cs, tess->offchip_layout);
 		radeon_emit(cs, tess->tcs_out_offsets);
-		radeon_emit(cs, tess->tcs_out_layout);
 	}
 
 	loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_TESS_EVAL, AC_UD_TES_OFFCHIP_LAYOUT);
-- 
2.14.3