radv: handle primitive id input into fragment shader with no geom shader

Fixes:
dEQP-VK.pipeline.framebuffer_attachment.no_attachments
dEQP-VK.pipeline.framebuffer_attachment.no_attachments_ms

Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

6 files changed, 65 insertions, 13 deletions

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index afef3fd28d4..f0ae7087379 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -5124,6 +5124,7 @@ si_llvm_init_export_args(struct nir_to_llvm_context *ctx,
 
 static void
 handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
+		       bool export_prim_id,
 		       struct ac_vs_output_info *outinfo)
 {
 	uint32_t param_count = 0;
@@ -5265,6 +5266,23 @@ handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
 		ac_build_export(&ctx->ac, &pos_args[i]);
 	}
 
+
+	if (export_prim_id) {
+		LLVMValueRef values[4];
+		target = V_008DFC_SQ_EXP_PARAM + param_count;
+		outinfo->vs_output_param_offset[VARYING_SLOT_PRIMITIVE_ID] = param_count;
+		param_count++;
+
+		values[0] = ctx->vs_prim_id;
+		ctx->shader_info->vs.vgpr_comp_cnt = MAX2(2,
+							  ctx->shader_info->vs.vgpr_comp_cnt);
+		for (unsigned j = 1; j < 4; j++)
+			values[j] = ctx->f32zero;
+		si_llvm_init_export_args(ctx, values, target, &args);
+		ac_build_export(&ctx->ac, &args);
+		outinfo->export_prim_id = true;
+	}
+
 	outinfo->pos_exports = num_pos_exports;
 	outinfo->param_exports = param_count;
 }
@@ -5700,7 +5718,8 @@ handle_shader_outputs_post(struct nir_to_llvm_context *ctx)
 		else if (ctx->options->key.vs.as_es)
 			handle_es_outputs_post(ctx, &ctx->shader_info->vs.es_info);
 		else
-			handle_vs_outputs_post(ctx, &ctx->shader_info->vs.outinfo);
+			handle_vs_outputs_post(ctx, ctx->options->key.vs.export_prim_id,
+					       &ctx->shader_info->vs.outinfo);
 		break;
 	case MESA_SHADER_FRAGMENT:
 		handle_fs_outputs_post(ctx);
@@ -5715,7 +5734,8 @@ handle_shader_outputs_post(struct nir_to_llvm_context *ctx)
 		if (ctx->options->key.tes.as_es)
 			handle_es_outputs_post(ctx, &ctx->shader_info->tes.es_info);
 		else
-			handle_vs_outputs_post(ctx, &ctx->shader_info->tes.outinfo);
+			handle_vs_outputs_post(ctx, ctx->options->key.tes.export_prim_id,
+					       &ctx->shader_info->tes.outinfo);
 		break;
 	default:
 		break;
@@ -6191,7 +6211,7 @@ ac_gs_copy_shader_emit(struct nir_to_llvm_context *ctx)
 		}
 		idx += slot_inc;
 	}
-	handle_vs_outputs_post(ctx, &ctx->shader_info->vs.outinfo);
+	handle_vs_outputs_post(ctx, false, &ctx->shader_info->vs.outinfo);
 }
 
 void ac_create_gs_copy_shader(LLVMTargetMachineRef tm,
diff --git a/src/amd/common/ac_nir_to_llvm.h b/src/amd/common/ac_nir_to_llvm.h
index 724d5e6b0fd..54d54891ac1 100644
--- a/src/amd/common/ac_nir_to_llvm.h
+++ b/src/amd/common/ac_nir_to_llvm.h
@@ -41,10 +41,12 @@ struct ac_vs_variant_key {
 	uint32_t instance_rate_inputs;
 	uint32_t as_es:1;
 	uint32_t as_ls:1;
+	uint32_t export_prim_id:1;
 };
 
 struct ac_tes_variant_key {
 	uint32_t as_es:1;
+	uint32_t export_prim_id:1;
 };
 
 struct ac_tcs_variant_key {
@@ -128,6 +130,7 @@ struct ac_vs_output_info {
 	bool writes_pointsize;
 	bool writes_layer;
 	bool writes_viewport_index;
+	bool export_prim_id;
 	uint32_t export_mask;
 	unsigned pos_exports;
 };
diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 9f36d982234..105384d315d 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -669,7 +669,7 @@ radv_emit_vertex_shader(struct radv_cmd_buffer *cmd_buffer,
 	else
 		radv_emit_hw_vs(cmd_buffer, pipeline, vs, &vs->info.vs.outinfo);
 
-	radeon_set_context_reg(cmd_buffer->cs, R_028A84_VGT_PRIMITIVEID_EN, 0);
+	radeon_set_context_reg(cmd_buffer->cs, R_028A84_VGT_PRIMITIVEID_EN, pipeline->graphics.vgt_primitiveid_en);
 }
 
 
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index e58ada68cc9..0c262799f0c 100644
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -605,11 +605,14 @@ radv_pipeline_compile(struct radv_pipeline *pipeline,
 }
 
 static union ac_shader_variant_key
-radv_compute_tes_key(bool as_es)
+radv_compute_tes_key(bool as_es, bool export_prim_id)
 {
 	union ac_shader_variant_key key;
 	memset(&key, 0, sizeof(key));
 	key.tes.as_es = as_es;
+	/* export prim id only happens when no geom shader */
+	if (!as_es)
+		key.tes.export_prim_id = export_prim_id;
 	return key;
 }
 
@@ -640,10 +643,12 @@ radv_tess_pipeline_compile(struct radv_pipeline *pipeline,
 	nir_shader *tes_nir, *tcs_nir;
 	void *tes_code = NULL, *tcs_code = NULL;
 	unsigned tes_code_size = 0, tcs_code_size = 0;
-	union ac_shader_variant_key tes_key = radv_compute_tes_key(radv_pipeline_has_gs(pipeline));
+	union ac_shader_variant_key tes_key;
 	union ac_shader_variant_key tcs_key;
 	bool dump = (pipeline->device->debug_flags & RADV_DEBUG_DUMP_SHADERS);
 
+	tes_key = radv_compute_tes_key(radv_pipeline_has_gs(pipeline),
+				       pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input);
 	if (tes_module->nir)
 		_mesa_sha1_compute(tes_module->nir->info.name,
 				   strlen(tes_module->nir->info.name),
@@ -1606,7 +1611,7 @@ radv_pipeline_init_dynamic_state(struct radv_pipeline *pipeline,
 }
 
 static union ac_shader_variant_key
-radv_compute_vs_key(const VkGraphicsPipelineCreateInfo *pCreateInfo, bool as_es, bool as_ls)
+radv_compute_vs_key(const VkGraphicsPipelineCreateInfo *pCreateInfo, bool as_es, bool as_ls, bool export_prim_id)
 {
 	union ac_shader_variant_key key;
 	const VkPipelineVertexInputStateCreateInfo *input_state =
@@ -1616,6 +1621,7 @@ radv_compute_vs_key(const VkGraphicsPipelineCreateInfo *pCreateInfo, bool as_es,
 	key.vs.instance_rate_inputs = 0;
 	key.vs.as_es = as_es;
 	key.vs.as_ls = as_ls;
+	key.vs.export_prim_id = export_prim_id;
 
 	for (unsigned i = 0; i < input_state->vertexAttributeDescriptionCount; ++i) {
 		unsigned binding;
@@ -1857,6 +1863,24 @@ static uint32_t si_vgt_gs_mode(struct radv_shader_variant *gs)
 	       S_028A40_GS_WRITE_OPTIMIZE(1);
 }
 
+static void calculate_vgt_gs_mode(struct radv_pipeline *pipeline)
+{
+	struct radv_shader_variant *vs;
+	vs = radv_pipeline_has_gs(pipeline) ? pipeline->gs_copy_shader : (radv_pipeline_has_tess(pipeline) ? pipeline->shaders[MESA_SHADER_TESS_EVAL] :  pipeline->shaders[MESA_SHADER_VERTEX]);
+
+	struct ac_vs_output_info *outinfo = &vs->info.vs.outinfo;
+
+	pipeline->graphics.vgt_primitiveid_en = false;
+	pipeline->graphics.vgt_gs_mode = 0;
+
+	if (radv_pipeline_has_gs(pipeline)) {
+		pipeline->graphics.vgt_gs_mode = si_vgt_gs_mode(pipeline->shaders[MESA_SHADER_GEOMETRY]);
+	} else if (outinfo->export_prim_id) {
+		pipeline->graphics.vgt_gs_mode = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
+		pipeline->graphics.vgt_primitiveid_en = true;
+	}
+}
+
 static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
 {
 	struct radv_shader_variant *vs;
@@ -2018,11 +2042,14 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
 	if (modules[MESA_SHADER_VERTEX]) {
 		bool as_es = false;
 		bool as_ls = false;
+		bool export_prim_id = false;
 		if (modules[MESA_SHADER_TESS_CTRL])
 			as_ls = true;
 		else if (modules[MESA_SHADER_GEOMETRY])
 			as_es = true;
-		union ac_shader_variant_key key = radv_compute_vs_key(pCreateInfo, as_es, as_ls);
+		else if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input)
+			export_prim_id = true;
+		union ac_shader_variant_key key = radv_compute_vs_key(pCreateInfo, as_es, as_ls, export_prim_id);
 
 		pipeline->shaders[MESA_SHADER_VERTEX] =
 			 radv_pipeline_compile(pipeline, cache, modules[MESA_SHADER_VERTEX],
@@ -2035,7 +2062,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
 	}
 
 	if (modules[MESA_SHADER_GEOMETRY]) {
-		union ac_shader_variant_key key = radv_compute_vs_key(pCreateInfo, false, false);
+		union ac_shader_variant_key key = radv_compute_vs_key(pCreateInfo, false, false, false);
 
 		pipeline->shaders[MESA_SHADER_GEOMETRY] =
 			 radv_pipeline_compile(pipeline, cache, modules[MESA_SHADER_GEOMETRY],
@@ -2045,10 +2072,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
 					       pipeline->layout, &key);
 
 		pipeline->active_stages |= mesa_to_vk_shader_stage(MESA_SHADER_GEOMETRY);
-
-		pipeline->graphics.vgt_gs_mode = si_vgt_gs_mode(pipeline->shaders[MESA_SHADER_GEOMETRY]);
-	} else
-		pipeline->graphics.vgt_gs_mode = 0;
+	}
 
 	if (modules[MESA_SHADER_TESS_EVAL]) {
 		assert(modules[MESA_SHADER_TESS_CTRL]);
@@ -2130,6 +2154,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
 		ps->info.fs.writes_z ? V_028710_SPI_SHADER_32_R :
 		V_028710_SPI_SHADER_ZERO;
 
+	calculate_vgt_gs_mode(pipeline);
 	calculate_pa_cl_vs_out_cntl(pipeline);
 	calculate_ps_inputs(pipeline);
 
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 87cb0a67fe7..630448d3b2a 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1071,6 +1071,7 @@ struct radv_pipeline {
 			unsigned prim;
 			unsigned gs_out;
 			uint32_t vgt_gs_mode;
+			bool vgt_primitiveid_en;
 			bool prim_restart_enable;
 			unsigned esgs_ring_size;
 			unsigned gsvs_ring_size;
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index 86e8e3e4986..bf08214c551 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -700,6 +700,9 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
 
 	multi_instances_smaller_than_primgroup = indirect_draw || (instanced_draw &&
 								   num_prims < primgroup_size);
+	if (cmd_buffer->state.pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input)
+		ia_switch_on_eoi = true;
+
 	if (radv_pipeline_has_tess(cmd_buffer->state.pipeline)) {
 		/* SWITCH_ON_EOI must be set if PrimID is used. */
 		if (cmd_buffer->state.pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.uses_prim_id ||