ac,radeonsi: add ac_shader_abi::emit_outputs for hardware VS shaders

3 files changed, 37 insertions, 11 deletions

diff --git a/src/amd/common/ac_shader_abi.h b/src/amd/common/ac_shader_abi.h
index a872948dd0..81fbc22bb4 100644
--- a/src/amd/common/ac_shader_abi.h
+++ b/src/amd/common/ac_shader_abi.h
@@ -42,6 +42,10 @@ struct ac_shader_abi {
 	 * driver_location.
 	 */
 	LLVMValueRef *inputs;
+
+	void (*emit_outputs)(struct ac_shader_abi *abi,
+			     unsigned max_outputs,
+			     LLVMValueRef *addrs);
 };
 
 #endif /* AC_SHADER_ABI_H */
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 0619c0f8fd..2d1a5808b9 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2998,15 +2998,18 @@ static void si_llvm_emit_gs_epilogue(struct lp_build_tgsi_context *bld_base)
 			 si_get_gs_wave_id(ctx));
 }
 
-static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
+static void si_llvm_emit_vs_epilogue(struct ac_shader_abi *abi,
+				     unsigned max_outputs,
+				     LLVMValueRef *addrs)
 {
-	struct si_shader_context *ctx = si_shader_context(bld_base);
+	struct si_shader_context *ctx = si_shader_context_from_abi(abi);
 	struct gallivm_state *gallivm = &ctx->gallivm;
 	struct tgsi_shader_info *info = &ctx->shader->selector->info;
 	struct si_shader_output_values *outputs = NULL;
 	int i,j;
 
 	assert(!ctx->shader->is_gs_copy_shader);
+	assert(info->num_outputs <= max_outputs);
 
 	outputs = MALLOC((info->num_outputs + 1) * sizeof(outputs[0]));
 
@@ -3037,7 +3040,7 @@ static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
 			}
 
 			for (j = 0; j < 4; j++) {
-				addr = ctx->outputs[i][j];
+				addr = addrs[4 * i + j];
 				val = LLVMBuildLoad(gallivm->builder, addr, "");
 				val = ac_build_clamp(&ctx->ac, val);
 				LLVMBuildStore(gallivm->builder, val, addr);
@@ -3055,7 +3058,7 @@ static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
 		for (j = 0; j < 4; j++) {
 			outputs[i].values[j] =
 				LLVMBuildLoad(gallivm->builder,
-					      ctx->outputs[i][j],
+					      addrs[4 * i + j],
 					      "");
 			outputs[i].vertex_stream[j] =
 				(info->output_streams[i] >> (2 * j)) & 3;
@@ -3069,8 +3072,8 @@ static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
 	if (ctx->shader->key.mono.u.vs_export_prim_id) {
 		outputs[i].semantic_name = TGSI_SEMANTIC_PRIMID;
 		outputs[i].semantic_index = 0;
-		outputs[i].values[0] = bitcast(bld_base, TGSI_TYPE_FLOAT,
-					       get_primitive_id(ctx, 0));
+		outputs[i].values[0] = LLVMBuildBitCast(gallivm->builder,
+				get_primitive_id(ctx, 0), ctx->f32, "");
 		for (j = 1; j < 4; j++)
 			outputs[i].values[j] = LLVMConstReal(ctx->f32, 0);
 
@@ -3079,10 +3082,18 @@ static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
 		i++;
 	}
 
-	si_llvm_export_vs(bld_base, outputs, i);
+	si_llvm_export_vs(&ctx->bld_base, outputs, i);
 	FREE(outputs);
 }
 
+static void si_tgsi_emit_epilogue(struct lp_build_tgsi_context *bld_base)
+{
+	struct si_shader_context *ctx = si_shader_context(bld_base);
+
+	ctx->abi.emit_outputs(&ctx->abi, RADEON_LLVM_MAX_OUTPUTS,
+			      &ctx->outputs[0][0]);
+}
+
 struct si_ps_exports {
 	unsigned num;
 	struct ac_export_args args[10];
@@ -5533,8 +5544,10 @@ static bool si_compile_tgsi_main(struct si_shader_context *ctx,
 			bld_base->emit_epilogue = si_llvm_emit_ls_epilogue;
 		else if (shader->key.as_es)
 			bld_base->emit_epilogue = si_llvm_emit_es_epilogue;
-		else
-			bld_base->emit_epilogue = si_llvm_emit_vs_epilogue;
+		else {
+			ctx->abi.emit_outputs = si_llvm_emit_vs_epilogue;
+			bld_base->emit_epilogue = si_tgsi_emit_epilogue;
+		}
 		break;
 	case PIPE_SHADER_TESS_CTRL:
 		bld_base->emit_fetch_funcs[TGSI_FILE_INPUT] = fetch_input_tcs;
@@ -5546,8 +5559,10 @@ static bool si_compile_tgsi_main(struct si_shader_context *ctx,
 		bld_base->emit_fetch_funcs[TGSI_FILE_INPUT] = fetch_input_tes;
 		if (shader->key.as_es)
 			bld_base->emit_epilogue = si_llvm_emit_es_epilogue;
-		else
-			bld_base->emit_epilogue = si_llvm_emit_vs_epilogue;
+		else {
+			ctx->abi.emit_outputs = si_llvm_emit_vs_epilogue;
+			bld_base->emit_epilogue = si_tgsi_emit_epilogue;
+		}
 		break;
 	case PIPE_SHADER_GEOMETRY:
 		bld_base->emit_fetch_funcs[TGSI_FILE_INPUT] = fetch_input_gs;
diff --git a/src/gallium/drivers/radeonsi/si_shader_internal.h b/src/gallium/drivers/radeonsi/si_shader_internal.h
index 2054a73610..c8dcd5c88e 100644
--- a/src/gallium/drivers/radeonsi/si_shader_internal.h
+++ b/src/gallium/drivers/radeonsi/si_shader_internal.h
@@ -242,6 +242,13 @@ si_shader_context(struct lp_build_tgsi_context *bld_base)
 	return (struct si_shader_context*)bld_base;
 }
 
+static inline struct si_shader_context *
+si_shader_context_from_abi(struct ac_shader_abi *abi)
+{
+	struct si_shader_context *ctx = NULL;
+	return container_of(abi, ctx, abi);
+}
+
 void si_llvm_add_attribute(LLVMValueRef F, const char *name, int value);
 
 LLVMTargetRef si_llvm_get_amdgpu_target(const char *triple);