[Mesa-dev] [PATCH 07/10] radeonsi: allow specifying simm16 of emit_waitcnt at call sites

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

The next commit will use this.

Cc: 13.0 12.0 <mesa-stable at lists.freedesktop.org>
---
 src/gallium/drivers/radeonsi/si_shader.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 145de9f..b914efb 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -3184,39 +3184,41 @@ static void build_tex_intrinsic(const struct lp_build_tgsi_action *action,
  * having side effects.
  */
 static void emit_optimization_barrier(struct si_shader_context *ctx)
 {
 	LLVMBuilderRef builder = ctx->gallivm.builder;
 	LLVMTypeRef ftype = LLVMFunctionType(ctx->voidt, NULL, 0, false);
 	LLVMValueRef inlineasm = LLVMConstInlineAsm(ftype, "", "", true, false);
 	LLVMBuildCall(builder, inlineasm, NULL, 0, "");
 }
 
-static void emit_waitcnt(struct si_shader_context *ctx)
+#define VM_CNT 0xf70
+
+static void emit_waitcnt(struct si_shader_context *ctx, unsigned simm16)
 {
 	struct gallivm_state *gallivm = &ctx->gallivm;
 	LLVMBuilderRef builder = gallivm->builder;
 	LLVMValueRef args[1] = {
-		lp_build_const_int32(gallivm, 0xf70)
+		lp_build_const_int32(gallivm, simm16)
 	};
 	lp_build_intrinsic(builder, "llvm.amdgcn.s.waitcnt",
 			   ctx->voidt, args, 1, 0);
 }
 
 static void membar_emit(
 		const struct lp_build_tgsi_action *action,
 		struct lp_build_tgsi_context *bld_base,
 		struct lp_build_emit_data *emit_data)
 {
 	struct si_shader_context *ctx = si_shader_context(bld_base);
 
-	emit_waitcnt(ctx);
+	emit_waitcnt(ctx, VM_CNT);
 }
 
 static LLVMValueRef
 shader_buffer_fetch_rsrc(struct si_shader_context *ctx,
 			 const struct tgsi_full_src_register *reg)
 {
 	LLVMValueRef index;
 	LLVMValueRef rsrc_ptr = LLVMGetParam(ctx->main_fn,
 					     SI_PARAM_SHADER_BUFFERS);
 
@@ -3606,21 +3608,21 @@ static void load_emit(
 	LLVMBuilderRef builder = gallivm->builder;
 	const struct tgsi_full_instruction * inst = emit_data->inst;
 	char intrinsic_name[64];
 
 	if (inst->Src[0].Register.File == TGSI_FILE_MEMORY) {
 		load_emit_memory(ctx, emit_data);
 		return;
 	}
 
 	if (inst->Memory.Qualifier & TGSI_MEMORY_VOLATILE)
-		emit_waitcnt(ctx);
+		emit_waitcnt(ctx, VM_CNT);
 
 	if (inst->Src[0].Register.File == TGSI_FILE_BUFFER) {
 		load_emit_buffer(ctx, emit_data);
 		return;
 	}
 
 	if (inst->Memory.Texture == TGSI_TEXTURE_BUFFER) {
 		emit_data->output[emit_data->chan] =
 			lp_build_intrinsic(
 				builder, "llvm.amdgcn.buffer.load.format.v4f32", emit_data->dst_type,
@@ -3815,21 +3817,21 @@ static void store_emit(
 	const struct tgsi_full_instruction * inst = emit_data->inst;
 	unsigned target = inst->Memory.Texture;
 	char intrinsic_name[64];
 
 	if (inst->Dst[0].Register.File == TGSI_FILE_MEMORY) {
 		store_emit_memory(ctx, emit_data);
 		return;
 	}
 
 	if (inst->Memory.Qualifier & TGSI_MEMORY_VOLATILE)
-		emit_waitcnt(ctx);
+		emit_waitcnt(ctx, VM_CNT);
 
 	if (inst->Dst[0].Register.File == TGSI_FILE_BUFFER) {
 		store_emit_buffer(ctx, emit_data);
 		return;
 	}
 
 	if (target == TGSI_TEXTURE_BUFFER) {
 		emit_data->output[emit_data->chan] = lp_build_intrinsic(
 			builder, "llvm.amdgcn.buffer.store.format.v4f32",
 			emit_data->dst_type, emit_data->args,
-- 
2.7.4