[Mesa-dev] [PATCH 14/14] R600/SI: inline V_ADD|SUB_F32 patterns

[Tom Stellard]

On Wed, Feb 20, 2013 at 06:47:00PM +0100, Christian König wrote:
> From: Christian König <christian.koenig at amd.com>
> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>  lib/Target/R600/SIInstructions.td |   14 +++++---------
>  1 file changed, 5 insertions(+), 9 deletions(-)
> 
> diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
> index 71ae33b..9ea26d2 100644
> --- a/lib/Target/R600/SIInstructions.td
> +++ b/lib/Target/R600/SIInstructions.td
> @@ -746,17 +746,13 @@ def : Pat <
>  defm V_READLANE_B32 : VOP2_32 <0x00000001, "V_READLANE_B32", []>;
>  defm V_WRITELANE_B32 : VOP2_32 <0x00000002, "V_WRITELANE_B32", []>;
>  
> -defm V_ADD_F32 : VOP2_32 <0x00000003, "V_ADD_F32", []>;
> -def : Pat <
> -  (f32 (fadd VSrc_32:$src0, VReg_32:$src1)),
> -  (V_ADD_F32_e32  VSrc_32:$src0, VReg_32:$src1)
> +defm V_ADD_F32 : VOP2_32 <0x00000003, "V_ADD_F32",
> +  [(set VReg_32:$dst, (fadd VSrc_32:$src0, VReg_32:$src1))]
>  >;
> -
> -defm V_SUB_F32 : VOP2_32 <0x00000004, "V_SUB_F32", []>;
> -def : Pat <
> -  (f32 (fsub VSrc_32:$src0, VReg_32:$src1)),
> -  (V_SUB_F32_e32  VSrc_32:$src0, VReg_32:$src1)
> +defm V_SUB_F32 : VOP2_32 <0x00000004, "V_SUB_F32",
> +  [(set VReg_32:$dst, (fsub VSrc_32:$src0, VReg_32:$src1))]
>  >;
> +

This looks good for now, but in the future it may be useful to simplify
all the VOP* and SOP* patterns by adding a helper that takes only an
SDPatternOperator as an argument and use that to build the pattern.
As an example, take a look at the R600_[123]OP_Helper classes in
R600Instructions.td

>  defm V_SUBREV_F32 : VOP2_32 <0x00000005, "V_SUBREV_F32", []>;
>  defm V_MAC_LEGACY_F32 : VOP2_32 <0x00000006, "V_MAC_LEGACY_F32", []>;
>  defm V_MUL_LEGACY_F32 : VOP2_32 <
> -- 
> 1.7.10.4
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits