[PATCH 1/1] R600: Enable vector fpow.

[Jan Vesely]

The OpenCL specs say: "The vector versions of the math functions operate
component-wise. The description is per-component."

Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
CC: Tom Stellard <thomas.stellard at amd.com>
---
Hi,

this patch enables vector pow. The OpenCL specs say that vector op is
per-component. LLVM pow intrinsic allows vectors too, but does not specify
behavior in vector case, so I assumed it was the same.

This patch fixes GEGL's "kernel_exposure" (operations/common/exposure.c).

I'm not sure about the test part, as I don't know what influences the order of
operations.

regards,
Jan

 lib/Target/R600/AMDGPUISelLowering.cpp |  1 +
 test/CodeGen/R600/llvm.pow.ll          | 29 +++++++++++++++++++++++++----
 2 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index e0ed721..f97a53f 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -185,6 +185,7 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
     setOperationAction(ISD::FABS, VT, Expand);
     setOperationAction(ISD::FADD, VT, Expand);
     setOperationAction(ISD::FDIV, VT, Expand);
+    setOperationAction(ISD::FPOW, VT, Expand);
     setOperationAction(ISD::FFLOOR, VT, Expand);
     setOperationAction(ISD::FTRUNC, VT, Expand);
     setOperationAction(ISD::FMUL, VT, Expand);
diff --git a/test/CodeGen/R600/llvm.pow.ll b/test/CodeGen/R600/llvm.pow.ll
index b587d2b..af84a1a 100644
--- a/test/CodeGen/R600/llvm.pow.ll
+++ b/test/CodeGen/R600/llvm.pow.ll
@@ -1,10 +1,11 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
-;CHECK: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: MUL NON-IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}}
-;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
+;CHECK: test1:
+;CHECK: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}},
+;CHECK-NEXT: MUL NON-IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}},
+;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}},
 
-define void @test(<4 x float> inreg %reg0) #0 {
+define void @test1(<4 x float> inreg %reg0) #0 {
    %r0 = extractelement <4 x float> %reg0, i32 0
    %r1 = extractelement <4 x float> %reg0, i32 1
    %r2 = call float @llvm.pow.f32( float %r0, float %r1)
@@ -13,7 +14,27 @@ define void @test(<4 x float> inreg %reg0) #0 {
    ret void
 }
 
+;CHECK: test2:
+;CHECK: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}},
+;CHECK-NEXT: MUL NON-IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}},
+;CHECK-NEXT: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}},
+;CHECK-NEXT: MUL NON-IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}},
+;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}},
+;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}},
+;CHECK-NEXT: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}},
+;CHECK-NEXT: MUL NON-IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}},
+;CHECK-NEXT: LOG_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}},
+;CHECK-NEXT: MUL NON-IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], PS}},
+;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}},
+;CHECK-NEXT: EXP_IEEE * T{{[0-9]+\.[XYZW], PV\.[XYZW]}},
+define void @test2(<4 x float> inreg %reg0, <4 x float> inreg %reg1) #0 {
+   %vec = call <4 x float> @llvm.pow.v4f32( <4 x float> %reg0, <4 x float> %reg1)
+   call void @llvm.R600.store.swizzle(<4 x float> %vec, i32 0, i32 0)
+   ret void
+}
+
 declare float @llvm.pow.f32(float ,float ) readonly
+declare <4 x float> @llvm.pow.v4f32(<4 x float> ,<4 x float> ) readonly
 declare void @llvm.R600.store.swizzle(<4 x float>, i32, i32)
 
 attributes #0 = { "ShaderType"="0" }
-- 
1.8.5.3