ac: add ac_build_readlane without optimization barrier

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

diff --git a/src/amd/llvm/ac_llvm_build.c b/src/amd/llvm/ac_llvm_build.c
index f789ff5..3640c7c 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.c
+++ b/src/amd/llvm/ac_llvm_build.c
@@ -3595,8 +3595,6 @@ _ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef l
        LLVMTypeRef type = LLVMTypeOf(src);
        LLVMValueRef result;
 
-       ac_build_optimization_barrier(ctx, &src);
-
        src = LLVMBuildZExt(ctx->builder, src, ctx->i32, "");
        if (lane)
                lane = LLVMBuildZExt(ctx->builder, lane, ctx->i32, "");
@@ -3613,13 +3611,17 @@ _ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef l
 
 /**
  * Builds the "llvm.amdgcn.readlane" or "llvm.amdgcn.readfirstlane" intrinsic.
+ *
+ * The optimization barrier is not needed if the value is the same in all lanes
+ * or if this is called in the outermost block.
+ *
  * @param ctx
  * @param src
  * @param lane - id of the lane or NULL for the first active lane
  * @return value of the lane
  */
-LLVMValueRef
-ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef lane)
+LLVMValueRef ac_build_readlane_no_opt_barrier(struct ac_llvm_context *ctx,
+                                             LLVMValueRef src, LLVMValueRef lane)
 {
        LLVMTypeRef src_type = LLVMTypeOf(src);
        src = ac_to_integer(ctx, src);
@@ -3649,6 +3651,14 @@ ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef la
 }
 
 LLVMValueRef
+ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef lane)
+{
+       ac_build_optimization_barrier(ctx, &src);
+
+       return ac_build_readlane_no_opt_barrier(ctx, src, lane);
+}
+
+LLVMValueRef
 ac_build_writelane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef value, LLVMValueRef lane)
 {
        return ac_build_intrinsic(ctx, "llvm.amdgcn.writelane", ctx->i32,

diff --git a/src/amd/llvm/ac_llvm_build.h b/src/amd/llvm/ac_llvm_build.h
index 56ec761..4be06a9 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.h
+++ b/src/amd/llvm/ac_llvm_build.h
@@ -655,6 +655,9 @@ void ac_apply_fmask_to_sample(struct ac_llvm_context *ac, LLVMValueRef fmask,
 LLVMValueRef
 ac_build_ds_swizzle(struct ac_llvm_context *ctx, LLVMValueRef src, unsigned mask);
 
+LLVMValueRef ac_build_readlane_no_opt_barrier(struct ac_llvm_context *ctx,
+                                             LLVMValueRef src, LLVMValueRef lane);
+
 LLVMValueRef
 ac_build_readlane(struct ac_llvm_context *ctx, LLVMValueRef src, LLVMValueRef lane);