[Mesa-dev] [PATCH 2/9] radeonsi: use V_BFE for extracting a sample index

Tom Stellard tom at stellard.net
Mon Mar 2 13:05:50 PST 2015


On Mon, Mar 02, 2015 at 12:54:16PM +0100, Marek Olšák wrote:
> From: Marek Olšák <marek.olsak at amd.com>
> 
> ---
>  src/gallium/drivers/radeonsi/si_shader.c | 22 ++++++++++++++++------
>  1 file changed, 16 insertions(+), 6 deletions(-)
> 
> diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
> index b0417ed..f125483 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.c
> +++ b/src/gallium/drivers/radeonsi/si_shader.c
> @@ -192,6 +192,20 @@ static int get_param_index(unsigned semantic_name, unsigned index,
>  }
>  
>  /**
> + * BitField Extract: ((value >> rshift) & ((1 << bitwidth) - 1))
> + */

Ideally, we would just add a pattern for this in the backend and emit generic
LLVM IR here.  This would also make it possible to share the code with llvmpipe.

I think the best place to do this would be in AMDGPUTargetLowering::performDAGCombine().

-Tom

> +static LLVMValueRef build_bfe(struct gallivm_state *gallivm,
> +			      LLVMValueRef value, LLVMValueRef rshift,
> +			      LLVMValueRef bitwidth)
> +{
> +	LLVMValueRef args[3] = {value, rshift, bitwidth};
> +
> +	return build_intrinsic(gallivm->builder, "llvm.AMDGPU.bfe.u32",
> +			       LLVMInt32TypeInContext(gallivm->context),
> +			       args, Elements(args), LLVMReadNoneAttribute);
> +}
> +
> +/**
>   * Build an LLVM bytecode indexed load using LLVMBuildGEP + LLVMBuildLoad.
>   * It's equivalent to doing a load from &base_ptr[index].
>   *
> @@ -1721,7 +1735,6 @@ static void tex_fetch_args(
>  
>  		/* Initialize some constants. */
>  		LLVMValueRef four = LLVMConstInt(uint_bld->elem_type, 4, 0);
> -		LLVMValueRef F = LLVMConstInt(uint_bld->elem_type, 0xF, 0);
>  
>  		/* Apply the formula. */
>  		LLVMValueRef fmask =
> @@ -1734,11 +1747,8 @@ static void tex_fetch_args(
>  		LLVMValueRef sample_index4 =
>  			LLVMBuildMul(gallivm->builder, address[sample_chan], four, "");
>  
> -		LLVMValueRef shifted_fmask =
> -			LLVMBuildLShr(gallivm->builder, fmask, sample_index4, "");
> -
> -		LLVMValueRef final_sample =
> -			LLVMBuildAnd(gallivm->builder, shifted_fmask, F, "");
> +		LLVMValueRef final_sample = build_bfe(gallivm, fmask,
> +						      sample_index4, four);
>  
>  		/* Don't rewrite the sample index if WORD1.DATA_FORMAT of the FMASK
>  		 * resource descriptor is 0 (invalid),
> -- 
> 2.1.0
> 
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-dev mailing list