[Mesa-dev] [PATCH 1/2] ac: add ac_build_fmin/fmax helpers

[Samuel Pitoiset]

Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>

On 01/04/2018 01:55 AM, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> ---
>   src/amd/common/ac_llvm_build.c | 32 ++++++++++++++++++--------------
>   src/amd/common/ac_llvm_build.h |  5 ++++-
>   2 files changed, 22 insertions(+), 15 deletions(-)
> 
> diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
> index 96ad715..7100e52 100644
> --- a/src/amd/common/ac_llvm_build.c
> +++ b/src/amd/common/ac_llvm_build.c
> @@ -1198,44 +1198,48 @@ ac_build_umsb(struct ac_llvm_context *ctx,
>   	msb = LLVMBuildSub(ctx->builder, LLVMConstInt(ctx->i32, 31, false),
>   			   msb, "");
>   
>   	/* check for zero */
>   	return LLVMBuildSelect(ctx->builder,
>   			       LLVMBuildICmp(ctx->builder, LLVMIntEQ, arg,
>   					     LLVMConstInt(ctx->i32, 0, 0), ""),
>   			       LLVMConstInt(ctx->i32, -1, true), msb, "");
>   }
>   
> +LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
> +			   LLVMValueRef b)
> +{
> +	LLVMValueRef args[2] = {a, b};
> +	return ac_build_intrinsic(ctx, "llvm.minnum.f32", ctx->f32, args, 2,
> +				  AC_FUNC_ATTR_READNONE);
> +}
> +
> +LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
> +			   LLVMValueRef b)
> +{
> +	LLVMValueRef args[2] = {a, b};
> +	return ac_build_intrinsic(ctx, "llvm.maxnum.f32", ctx->f32, args, 2,
> +				  AC_FUNC_ATTR_READNONE);
> +}
> +
>   LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a,
>   			   LLVMValueRef b)
>   {
>   	LLVMValueRef cmp = LLVMBuildICmp(ctx->builder, LLVMIntULE, a, b, "");
>   	return LLVMBuildSelect(ctx->builder, cmp, a, b, "");
>   }
>   
>   LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value)
>   {
>   	if (HAVE_LLVM >= 0x0500) {
> -		LLVMValueRef max[2] = {
> -			value,
> -			LLVMConstReal(ctx->f32, 0),
> -		};
> -		LLVMValueRef min[2] = {
> -			LLVMConstReal(ctx->f32, 1),
> -		};
> -
> -		min[1] = ac_build_intrinsic(ctx, "llvm.maxnum.f32",
> -					    ctx->f32, max, 2,
> -					    AC_FUNC_ATTR_READNONE);
> -		return ac_build_intrinsic(ctx, "llvm.minnum.f32",
> -					  ctx->f32, min, 2,
> -					  AC_FUNC_ATTR_READNONE);
> +		return ac_build_fmin(ctx, ac_build_fmax(ctx, value, ctx->f32_0),
> +				     ctx->f32_1);
>   	}
>   
>   	LLVMValueRef args[3] = {
>   		value,
>   		LLVMConstReal(ctx->f32, 0),
>   		LLVMConstReal(ctx->f32, 1),
>   	};
>   
>   	return ac_build_intrinsic(ctx, "llvm.AMDGPU.clamp.", ctx->f32, args, 3,
>   				  AC_FUNC_ATTR_READNONE |
> diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
> index 3e0285b..0deb5b5 100644
> --- a/src/amd/common/ac_llvm_build.h
> +++ b/src/amd/common/ac_llvm_build.h
> @@ -234,21 +234,24 @@ void ac_build_sendmsg(struct ac_llvm_context *ctx,
>   		      uint32_t msg,
>   		      LLVMValueRef wave_id);
>   
>   LLVMValueRef ac_build_imsb(struct ac_llvm_context *ctx,
>   			   LLVMValueRef arg,
>   			   LLVMTypeRef dst_type);
>   
>   LLVMValueRef ac_build_umsb(struct ac_llvm_context *ctx,
>   			  LLVMValueRef arg,
>   			  LLVMTypeRef dst_type);
> -
> +LLVMValueRef ac_build_fmin(struct ac_llvm_context *ctx, LLVMValueRef a,
> +			   LLVMValueRef b);
> +LLVMValueRef ac_build_fmax(struct ac_llvm_context *ctx, LLVMValueRef a,
> +			   LLVMValueRef b);
>   LLVMValueRef ac_build_umin(struct ac_llvm_context *ctx, LLVMValueRef a, LLVMValueRef b);
>   LLVMValueRef ac_build_clamp(struct ac_llvm_context *ctx, LLVMValueRef value);
>   
>   struct ac_export_args {
>   	LLVMValueRef out[4];
>           unsigned target;
>           unsigned enabled_channels;
>           bool compr;
>           bool done;
>           bool valid_mask;
>