[Mesa-dev] [PATCH 12/14] ac/radv: migrate lds size calculations to shader gen.

[Dave Airlie]

From: Dave Airlie <airlied at redhat.com>

This moves the lds_size calcs into the shader so we have all
the size stuff in one file.

Signed-off-by: Dave Airlie <airlied at redhat.com>
---
 src/amd/common/ac_nir_to_llvm.c | 33 +++++++++++++++++++++++++++++++++
 src/amd/common/ac_nir_to_llvm.h |  1 +
 src/amd/vulkan/radv_pipeline.c  | 30 ++++--------------------------
 3 files changed, 38 insertions(+), 26 deletions(-)

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index c235c5314be..1cf181ddeba 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -357,6 +357,38 @@ get_tcs_num_patches(struct radv_shader_context *ctx)
 	return num_patches;
 }
 
+static unsigned
+calculate_tess_lds_size(struct radv_shader_context *ctx)
+{
+	unsigned num_tcs_input_cp = ctx->options->key.tcs.input_vertices;
+	unsigned num_tcs_output_cp;
+	unsigned num_tcs_outputs, num_tcs_patch_outputs;
+	unsigned input_vertex_size, output_vertex_size;
+	unsigned input_patch_size, output_patch_size;
+	unsigned pervertex_output_patch_size;
+	unsigned output_patch0_offset;
+	unsigned num_patches;
+	unsigned lds_size;
+
+	num_tcs_output_cp = ctx->tcs_vertices_per_patch;
+	num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
+	num_tcs_patch_outputs = util_last_bit64(ctx->shader_info->info.tcs.patch_outputs_written);
+
+	input_vertex_size = ctx->tcs_num_inputs * 16;
+	output_vertex_size = num_tcs_outputs * 16;
+
+	input_patch_size = num_tcs_input_cp * input_vertex_size;
+
+	pervertex_output_patch_size = num_tcs_output_cp * output_vertex_size;
+	output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16;
+
+	num_patches = ctx->tcs_num_patches;
+	output_patch0_offset = input_patch_size * num_patches;
+
+	lds_size = output_patch0_offset + output_patch_size * num_patches;
+	return lds_size;
+}
+
 /* Tessellation shaders pass outputs to the next shader using LDS.
  *
  * LS outputs = TCS inputs
@@ -6981,6 +7013,7 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
 				shaders[i]->info.gs.vertices_out;
 		} 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 62ce72fb7d2..f4825ef4cd5 100644
--- a/src/amd/common/ac_nir_to_llvm.h
+++ b/src/amd/common/ac_nir_to_llvm.h
@@ -199,6 +199,7 @@ struct ac_shader_variant_info {
 			unsigned tcs_vertices_out;
 			/* Which outputs are actually written */
 			uint32_t num_patches;
+			uint32_t lds_size;
 		} tcs;
 		struct {
 			struct ac_vs_output_info outinfo;
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index 195daf6abde..e9a9ae975b1 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -1306,39 +1306,17 @@ static struct radv_tessellation_state
 calculate_tess_state(struct radv_pipeline *pipeline,
 		     const VkGraphicsPipelineCreateInfo *pCreateInfo)
 {
-	unsigned num_tcs_input_cp = pCreateInfo->pTessellationState->patchControlPoints;
-	unsigned num_tcs_output_cp, num_tcs_inputs, num_tcs_outputs;
-	unsigned num_tcs_patch_outputs;
-	unsigned input_vertex_size, output_vertex_size, pervertex_output_patch_size;
-	unsigned input_patch_size, output_patch_size, output_patch0_offset;
+	unsigned num_tcs_input_cp;
+	unsigned num_tcs_output_cp;
 	unsigned lds_size;
 	unsigned num_patches;
 	struct radv_tessellation_state tess = {0};
 
-	/* This calculates how shader inputs and outputs among VS, TCS, and TES
-	 * are laid out in LDS. */
-	num_tcs_inputs = util_last_bit64(radv_get_vertex_shader(pipeline)->info.info.vs.ls_outputs_written);
-	num_tcs_outputs = util_last_bit64(pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.info.tcs.outputs_written); //tcs->outputs_written
+	num_tcs_input_cp = pCreateInfo->pTessellationState->patchControlPoints;
 	num_tcs_output_cp = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.tcs_vertices_out; //TCS VERTICES OUT
-	num_tcs_patch_outputs = util_last_bit64(pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.info.tcs.patch_outputs_written);
-
-	/* Ensure that we only need one wave per SIMD so we don't need to check
-	 * resource usage. Also ensures that the number of tcs in and out
-	 * vertices per threadgroup are at most 256.
-	 */
-	input_vertex_size = num_tcs_inputs * 16;
-	output_vertex_size = num_tcs_outputs * 16;
-
-	input_patch_size = num_tcs_input_cp * input_vertex_size;
-
-	pervertex_output_patch_size = num_tcs_output_cp * output_vertex_size;
-	output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16;
-
 	num_patches = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.num_patches;
-	output_patch0_offset = input_patch_size * num_patches;
-	/*	perpatch_output_offset = output_patch0_offset + pervertex_output_patch_size;*/
 
-	lds_size = output_patch0_offset + output_patch_size * num_patches;
+	lds_size = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.lds_size;
 
 	if (pipeline->device->physical_device->rad_info.chip_class >= CIK) {
 		assert(lds_size <= 65536);
-- 
2.14.3