[Mesa-dev] [PATCH] R600: Expand vector FSQRT ops

[Tom Stellard]

From: Tom Stellard <thomas.stellard at amd.com>

---
 lib/Target/R600/AMDGPUISelLowering.cpp |  1 +
 test/CodeGen/R600/llvm.sqrt.ll         | 54 ++++++++++++++++++++++++++++++++++
 2 files changed, 55 insertions(+)
 create mode 100644 test/CodeGen/R600/llvm.sqrt.ll

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 91d85d3..52dd010 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -181,6 +181,7 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
     setOperationAction(ISD::FFLOOR, VT, Expand);
     setOperationAction(ISD::FMUL, VT, Expand);
     setOperationAction(ISD::FRINT, VT, Expand);
+    setOperationAction(ISD::FSQRT, VT, Expand);
     setOperationAction(ISD::FSUB, VT, Expand);
   }
 }
diff --git a/test/CodeGen/R600/llvm.sqrt.ll b/test/CodeGen/R600/llvm.sqrt.ll
new file mode 100644
index 0000000..0d0d186
--- /dev/null
+++ b/test/CodeGen/R600/llvm.sqrt.ll
@@ -0,0 +1,54 @@
+; RUN: llc < %s -march=r600 --mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
+; RUN: llc < %s -march=r600 --mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
+
+; R600-CHECK-LABEL: @sqrt_f32
+; R600-CHECK: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[2].Z
+; R600-CHECK: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[2].Z, PS
+; SI-CHECK-LABEL: @sqrt_f32
+; SI-CHECK: V_SQRT_F32_e32
+define void @sqrt_f32(float addrspace(1)* %out, float %in) {
+entry:
+  %0 = call float @llvm.sqrt.f32(float %in)
+  store float %0, float addrspace(1)* %out
+  ret void
+}
+
+; R600-CHECK-LABEL: @sqrt_v2f32
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[2].W
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[2].W, PS
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].X
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].X, PS
+; SI-CHECK-LABEL: @sqrt_v2f32
+; SI-CHECK: V_SQRT_F32_e32
+; SI-CHECK: V_SQRT_F32_e32
+define void @sqrt_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %in) {
+entry:
+  %0 = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %in)
+  store <2 x float> %0, <2 x float> addrspace(1)* %out
+  ret void
+}
+
+; R600-CHECK-LABEL: @sqrt_v4f32
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].Y
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].Y, PS
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].Z
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].Z, PS
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].W
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].W, PS
+; R600-CHECK-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[4].X
+; R600-CHECK-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[4].X, PS
+; SI-CHECK-LABEL: @sqrt_v4f32
+; SI-CHECK: V_SQRT_F32_e32
+; SI-CHECK: V_SQRT_F32_e32
+; SI-CHECK: V_SQRT_F32_e32
+; SI-CHECK: V_SQRT_F32_e32
+define void @sqrt_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %in) {
+entry:
+  %0 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %in)
+  store <4 x float> %0, <4 x float> addrspace(1)* %out
+  ret void
+}
+
+declare float @llvm.sqrt.f32(float %in)
+declare <2 x float> @llvm.sqrt.v2f32(<2 x float> %in)
+declare <4 x float> @llvm.sqrt.v4f32(<4 x float> %in)
-- 
1.7.11.4