[llvm] r213882 - R600: Add FMA instructions for Evergreen
Matt Arsenault
Matthew.Arsenault at amd.com
Thu Jul 24 10:41:01 PDT 2014
Author: arsenm
Date: Thu Jul 24 12:41:01 2014
New Revision: 213882
URL: http://llvm.org/viewvc/llvm-project?rev=213882&view=rev
Log:
R600: Add FMA instructions for Evergreen
Added:
llvm/trunk/test/CodeGen/R600/fma.f64.ll
- copied, changed from r213877, llvm/trunk/test/CodeGen/R600/fma.ll
Modified:
llvm/trunk/lib/Target/R600/EvergreenInstructions.td
llvm/trunk/lib/Target/R600/R600Instructions.td
llvm/trunk/test/CodeGen/R600/fma.ll
Modified: llvm/trunk/lib/Target/R600/EvergreenInstructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/EvergreenInstructions.td?rev=213882&r1=213881&r2=213882&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/EvergreenInstructions.td (original)
+++ llvm/trunk/lib/Target/R600/EvergreenInstructions.td Thu Jul 24 12:41:01 2014
@@ -257,6 +257,12 @@ def VTX_READ_GLOBAL_128_eg : VTX_READ_12
let Predicates = [isEGorCayman] in {
+// Should be predicated on FeatureFP64
+// 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
@@ -313,6 +319,7 @@ def BIT_ALIGN_INT_eg : R600_3OP <0xC, "B
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>;
Modified: llvm/trunk/lib/Target/R600/R600Instructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/R600Instructions.td?rev=213882&r1=213881&r2=213882&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/R600Instructions.td (original)
+++ llvm/trunk/lib/Target/R600/R600Instructions.td Thu Jul 24 12:41:01 2014
@@ -915,6 +915,11 @@ class MULADD_IEEE_Common <bits<5> inst>
[(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))]
Copied: llvm/trunk/test/CodeGen/R600/fma.f64.ll (from r213877, llvm/trunk/test/CodeGen/R600/fma.ll)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/fma.f64.ll?p2=llvm/trunk/test/CodeGen/R600/fma.f64.ll&p1=llvm/trunk/test/CodeGen/R600/fma.ll&r1=213877&r2=213882&rev=213882&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/R600/fma.ll (original)
+++ llvm/trunk/test/CodeGen/R600/fma.f64.ll Thu Jul 24 12:41:01 2014
@@ -1,52 +1,9 @@
; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -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,
- float addrspace(1)* %in2, float addrspace(1)* %in3) {
- %r0 = load float addrspace(1)* %in1
- %r1 = load float addrspace(1)* %in2
- %r2 = load float addrspace(1)* %in3
- %r3 = tail call float @llvm.fma.f32(float %r0, float %r1, float %r2)
- store float %r3, float addrspace(1)* %out
- ret void
-}
-
-; FUNC-LABEL: @fma_v2f32
-; SI: V_FMA_F32
-; SI: V_FMA_F32
-define void @fma_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in1,
- <2 x float> addrspace(1)* %in2, <2 x float> addrspace(1)* %in3) {
- %r0 = load <2 x float> addrspace(1)* %in1
- %r1 = load <2 x float> addrspace(1)* %in2
- %r2 = load <2 x float> addrspace(1)* %in3
- %r3 = tail call <2 x float> @llvm.fma.v2f32(<2 x float> %r0, <2 x float> %r1, <2 x float> %r2)
- store <2 x float> %r3, <2 x float> addrspace(1)* %out
- ret void
-}
-
-; FUNC-LABEL: @fma_v4f32
-; SI: V_FMA_F32
-; SI: V_FMA_F32
-; SI: V_FMA_F32
-; SI: V_FMA_F32
-define void @fma_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in1,
- <4 x float> addrspace(1)* %in2, <4 x float> addrspace(1)* %in3) {
- %r0 = load <4 x float> addrspace(1)* %in1
- %r1 = load <4 x float> addrspace(1)* %in2
- %r2 = load <4 x float> addrspace(1)* %in3
- %r3 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %r0, <4 x float> %r1, <4 x float> %r2)
- 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\[[0-9]+:[0-9]+\]}}
Modified: llvm/trunk/test/CodeGen/R600/fma.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/fma.ll?rev=213882&r1=213881&r2=213882&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/R600/fma.ll (original)
+++ llvm/trunk/test/CodeGen/R600/fma.ll Thu Jul 24 12:41:01 2014
@@ -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 @@ define void @fma_v4f32(<4 x float> addrs
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
-}
More information about the llvm-commits
mailing list