[PATCH] R600: Add FMA instruction
Matt Arsenault
arsenm2 at gmail.com
Thu Jul 24 00:16:24 PDT 2014
On Jul 14, 2014, at 12:30 PM, Tom Stellard <tom at stellard.net> wrote:
> On Thu, Jun 26, 2014 at 01:49:53AM +0000, Matt Arsenault wrote:
>> http://reviews.llvm.org/D4307
>>
>> Files:
>> lib/Target/R600/EvergreenInstructions.td
>> lib/Target/R600/R600Instructions.td
>> test/CodeGen/R600/fma.f64.ll
>> test/CodeGen/R600/fma.ll
>
> The FMA instructions need to be predicated on FeatureFP64, since not all
> Evergreen support doubles.
>
> -Tom
>
It doesn’t really matter now since it’s commented out anyway, since it’s broken along with the rest of f64 on Evergreen. Adding that to the Predicates list also requires awkwardly using something like let Predicates = [isEG, FeatureFP64] since for some reason Evergreen doesn’t seem to be using the SubtargetPredicate stuff
>> Index: lib/Target/R600/EvergreenInstructions.td
>> ===================================================================
>> --- lib/Target/R600/EvergreenInstructions.td
>> +++ lib/Target/R600/EvergreenInstructions.td
>> @@ -256,6 +256,11 @@
>>
>> let Predicates = [isEGorCayman] in {
>>
>> +// def FMA_64 : R600_3OP <
>> +// 0xA, "FMA_64",
>> +// [(set f64:$dst, (fma f64:$src0, f64:$src1, f64:$src2))]
>> +// >;
>> +
>> // BFE_UINT - bit_extract, an optimization for mask and shift
>> // Src0 = Input
>> // Src1 = Offset
>> @@ -312,6 +317,7 @@
>> def : ROTRPattern <BIT_ALIGN_INT_eg>;
>> def MULADD_eg : MULADD_Common<0x14>;
>> def MULADD_IEEE_eg : MULADD_IEEE_Common<0x18>;
>> +def FMA_eg : FMA_Common<0x7>;
>> def ASHR_eg : ASHR_Common<0x15>;
>> def LSHR_eg : LSHR_Common<0x16>;
>> def LSHL_eg : LSHL_Common<0x17>;
>> Index: lib/Target/R600/R600Instructions.td
>> ===================================================================
>> --- lib/Target/R600/R600Instructions.td
>> +++ lib/Target/R600/R600Instructions.td
>> @@ -915,6 +915,11 @@
>> [(set f32:$dst, (fadd (fmul f32:$src0, f32:$src1), f32:$src2))]
>>> ;
>>
>> +class FMA_Common <bits<5> inst> : R600_3OP <
>> + inst, "FMA",
>> + [(set f32:$dst, (fma f32:$src0, f32:$src1, f32:$src2))]
>> +>;
>> +
>> class CNDE_Common <bits<5> inst> : R600_3OP <
>> inst, "CNDE",
>> [(set f32:$dst, (selectcc f32:$src0, FP_ZERO, f32:$src1, f32:$src2, COND_OEQ))]
>> Index: test/CodeGen/R600/fma.f64.ll
>> ===================================================================
>> --- /dev/null
>> +++ test/CodeGen/R600/fma.f64.ll
>> @@ -0,0 +1,46 @@
>> +; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
>> +
>> +declare double @llvm.fma.f64(double, double, double) nounwind readnone
>> +declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
>> +declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
>> +
>> +
>> +; FUNC-LABEL: @fma_f64
>> +; SI: V_FMA_F64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\]}}
>> +define void @fma_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
>> + double addrspace(1)* %in2, double addrspace(1)* %in3) {
>> + %r0 = load double addrspace(1)* %in1
>> + %r1 = load double addrspace(1)* %in2
>> + %r2 = load double addrspace(1)* %in3
>> + %r3 = tail call double @llvm.fma.f64(double %r0, double %r1, double %r2)
>> + store double %r3, double addrspace(1)* %out
>> + ret void
>> +}
>> +
>> +; FUNC-LABEL: @fma_v2f64
>> +; SI: V_FMA_F64
>> +; SI: V_FMA_F64
>> +define void @fma_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in1,
>> + <2 x double> addrspace(1)* %in2, <2 x double> addrspace(1)* %in3) {
>> + %r0 = load <2 x double> addrspace(1)* %in1
>> + %r1 = load <2 x double> addrspace(1)* %in2
>> + %r2 = load <2 x double> addrspace(1)* %in3
>> + %r3 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %r0, <2 x double> %r1, <2 x double> %r2)
>> + store <2 x double> %r3, <2 x double> addrspace(1)* %out
>> + ret void
>> +}
>> +
>> +; FUNC-LABEL: @fma_v4f64
>> +; SI: V_FMA_F64
>> +; SI: V_FMA_F64
>> +; SI: V_FMA_F64
>> +; SI: V_FMA_F64
>> +define void @fma_v4f64(<4 x double> addrspace(1)* %out, <4 x double> addrspace(1)* %in1,
>> + <4 x double> addrspace(1)* %in2, <4 x double> addrspace(1)* %in3) {
>> + %r0 = load <4 x double> addrspace(1)* %in1
>> + %r1 = load <4 x double> addrspace(1)* %in2
>> + %r2 = load <4 x double> addrspace(1)* %in3
>> + %r3 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %r0, <4 x double> %r1, <4 x double> %r2)
>> + store <4 x double> %r3, <4 x double> addrspace(1)* %out
>> + ret void
>> +}
>> Index: test/CodeGen/R600/fma.ll
>> ===================================================================
>> --- test/CodeGen/R600/fma.ll
>> +++ test/CodeGen/R600/fma.ll
>> @@ -1,13 +1,10 @@
>> ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
>> +; RUN: llc -march=r600 -mcpu=redwood -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
>>
>> declare float @llvm.fma.f32(float, float, float) nounwind readnone
>> declare <2 x float> @llvm.fma.v2f32(<2 x float>, <2 x float>, <2 x float>) nounwind readnone
>> declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone
>>
>> -declare double @llvm.fma.f64(double, double, double) nounwind readnone
>> -declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
>> -declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) nounwind readnone
>> -
>> ; FUNC-LABEL: @fma_f32
>> ; SI: V_FMA_F32 {{v[0-9]+, v[0-9]+, v[0-9]+, v[0-9]+}}
>> define void @fma_f32(float addrspace(1)* %out, float addrspace(1)* %in1,
>> @@ -47,43 +44,3 @@
>> store <4 x float> %r3, <4 x float> addrspace(1)* %out
>> ret void
>> }
>> -
>> -; FUNC-LABEL: @fma_f64
>> -; SI: V_FMA_F64 {{v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\], v\[[0-9]+:[0-9]+\]}}
>> -define void @fma_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
>> - double addrspace(1)* %in2, double addrspace(1)* %in3) {
>> - %r0 = load double addrspace(1)* %in1
>> - %r1 = load double addrspace(1)* %in2
>> - %r2 = load double addrspace(1)* %in3
>> - %r3 = tail call double @llvm.fma.f64(double %r0, double %r1, double %r2)
>> - store double %r3, double addrspace(1)* %out
>> - ret void
>> -}
>> -
>> -; FUNC-LABEL: @fma_v2f64
>> -; SI: V_FMA_F64
>> -; SI: V_FMA_F64
>> -define void @fma_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in1,
>> - <2 x double> addrspace(1)* %in2, <2 x double> addrspace(1)* %in3) {
>> - %r0 = load <2 x double> addrspace(1)* %in1
>> - %r1 = load <2 x double> addrspace(1)* %in2
>> - %r2 = load <2 x double> addrspace(1)* %in3
>> - %r3 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %r0, <2 x double> %r1, <2 x double> %r2)
>> - store <2 x double> %r3, <2 x double> addrspace(1)* %out
>> - ret void
>> -}
>> -
>> -; FUNC-LABEL: @fma_v4f64
>> -; SI: V_FMA_F64
>> -; SI: V_FMA_F64
>> -; SI: V_FMA_F64
>> -; SI: V_FMA_F64
>> -define void @fma_v4f64(<4 x double> addrspace(1)* %out, <4 x double> addrspace(1)* %in1,
>> - <4 x double> addrspace(1)* %in2, <4 x double> addrspace(1)* %in3) {
>> - %r0 = load <4 x double> addrspace(1)* %in1
>> - %r1 = load <4 x double> addrspace(1)* %in2
>> - %r2 = load <4 x double> addrspace(1)* %in3
>> - %r3 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %r0, <4 x double> %r1, <4 x double> %r2)
>> - store <4 x double> %r3, <4 x double> addrspace(1)* %out
>> - ret void
>> -}
>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
More information about the llvm-commits
mailing list