R600: Expand vector float operations for both SI and R600

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188596 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 128 insertions, 88 deletions

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 746c4794d1..25b1e54f11 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -115,14 +115,14 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::VSELECT, MVT::v2f32, Expand);
   setOperationAction(ISD::VSELECT, MVT::v4f32, Expand);
 
-  static const int types[] = {
+  static const int IntTypes[] = {
     (int)MVT::v2i32,
     (int)MVT::v4i32
   };
-  const size_t NumTypes = array_lengthof(types);
+  const size_t NumIntTypes = array_lengthof(IntTypes);
 
-  for (unsigned int x  = 0; x < NumTypes; ++x) {
-    MVT::SimpleValueType VT = (MVT::SimpleValueType)types[x];
+  for (unsigned int x  = 0; x < NumIntTypes; ++x) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)IntTypes[x];
     //Expand the following operations for the current type by default
     setOperationAction(ISD::ADD,  VT, Expand);
     setOperationAction(ISD::AND,  VT, Expand);
@@ -141,6 +141,20 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
     setOperationAction(ISD::VSELECT, VT, Expand);
     setOperationAction(ISD::XOR,  VT, Expand);
   }
+
+  static const int FloatTypes[] = {
+    (int)MVT::v2f32,
+    (int)MVT::v4f32
+  };
+  const size_t NumFloatTypes = array_lengthof(FloatTypes);
+
+  for (unsigned int x = 0; x < NumFloatTypes; ++x) {
+    MVT::SimpleValueType VT = (MVT::SimpleValueType)FloatTypes[x];
+    setOperationAction(ISD::FADD, VT, Expand);
+    setOperationAction(ISD::FDIV, VT, Expand);
+    setOperationAction(ISD::FMUL, VT, Expand);
+    setOperationAction(ISD::FSUB, VT, Expand);
+  }
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
index e10af2b16b..b822431fab 100644
--- a/lib/Target/R600/R600ISelLowering.cpp
+++ b/lib/Target/R600/R600ISelLowering.cpp
@@ -38,15 +38,6 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
 
   computeRegisterProperties();
 
-  setOperationAction(ISD::FADD, MVT::v4f32, Expand);
-  setOperationAction(ISD::FADD, MVT::v2f32, Expand);
-  setOperationAction(ISD::FMUL, MVT::v4f32, Expand);
-  setOperationAction(ISD::FMUL, MVT::v2f32, Expand);
-  setOperationAction(ISD::FDIV, MVT::v4f32, Expand);
-  setOperationAction(ISD::FDIV, MVT::v2f32, Expand);
-  setOperationAction(ISD::FSUB, MVT::v4f32, Expand);
-  setOperationAction(ISD::FSUB, MVT::v2f32, Expand);
-
   setOperationAction(ISD::FCOS, MVT::f32, Custom);
   setOperationAction(ISD::FSIN, MVT::f32, Custom);
 
diff --git a/test/CodeGen/R600/fadd.ll b/test/CodeGen/R600/fadd.ll
index 2716958e50..6d459679d4 100644
--- a/test/CodeGen/R600/fadd.ll
+++ b/test/CodeGen/R600/fadd.ll
@@ -1,23 +1,23 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; 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
 
-; CHECK: @fadd_f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
-define void @fadd_f32() {
-   %r0 = call float @llvm.R600.load.input(i32 0)
-   %r1 = call float @llvm.R600.load.input(i32 1)
-   %r2 = fadd float %r0, %r1
-   call void @llvm.AMDGPU.store.output(float %r2, i32 0)
+; R600-CHECK: @fadd_f32
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW]}}, KC0[2].Z, KC0[2].W
+; SI-CHECK: @fadd_f32
+; SI-CHECK: V_ADD_F32
+define void @fadd_f32(float addrspace(1)* %out, float %a, float %b) {
+entry:
+   %0 = fadd float %a, %b
+   store float %0, float addrspace(1)* %out
    ret void
 }
 
-declare float @llvm.R600.load.input(i32) readnone
-
-declare void @llvm.AMDGPU.store.output(float, i32)
-
-; CHECK: @fadd_v2f32
-; CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[3].X, KC0[3].Z
-; CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
+; R600-CHECK: @fadd_v2f32
+; R600-CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[3].X, KC0[3].Z
+; R600-CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
+; SI-CHECK: @fadd_v2f32
+; SI-CHECK: V_ADD_F32
+; SI-CHECK: V_ADD_F32
 define void @fadd_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
 entry:
   %0 = fadd <2 x float> %a, %b
@@ -25,12 +25,16 @@ entry:
   ret void
 }
 
-; CHECK: @fadd_v4f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
+; R600-CHECK: @fadd_v4f32
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; SI-CHECK: @fadd_v4f32
+; SI-CHECK: V_ADD_F32
+; SI-CHECK: V_ADD_F32
+; SI-CHECK: V_ADD_F32
+; SI-CHECK: V_ADD_F32
 define void @fadd_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
   %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
   %a = load <4 x float> addrspace(1) * %in
diff --git a/test/CodeGen/R600/fdiv.ll b/test/CodeGen/R600/fdiv.ll
index 6798eacfa3..84e9f67736 100644
--- a/test/CodeGen/R600/fdiv.ll
+++ b/test/CodeGen/R600/fdiv.ll
@@ -1,14 +1,20 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; 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
 
 ; These tests check that fdiv is expanded correctly and also test that the
 ; scheduler is scheduling the RECIP_IEEE and MUL_IEEE instructions in separate
 ; instruction groups.
 
-; CHECK: @fdiv_v2f32
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Z
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Y
-; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW]}}, KC0[3].X, PS
-; CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[2].W, PS
+; R600-CHECK: @fdiv_v2f32
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Z
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Y
+; R600-CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW]}}, KC0[3].X, PS
+; R600-CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[2].W, PS
+; SI-CHECK: @fdiv_v2f32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
 define void @fdiv_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
 entry:
   %0 = fdiv <2 x float> %a, %b
@@ -16,16 +22,24 @@ entry:
   ret void
 }
 
-; CHECK: @fdiv_v4f32
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
-; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
-; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
-; CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
-
+; R600-CHECK: @fdiv_v4f32
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; R600-CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; R600-CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; R600-CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; SI-CHECK: @fdiv_v4f32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
+; SI-CHECK-DAG: V_RCP_F32
+; SI-CHECK-DAG: V_MUL_F32
 define void @fdiv_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
   %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
   %a = load <4 x float> addrspace(1) * %in
diff --git a/test/CodeGen/R600/fmul.ll b/test/CodeGen/R600/fmul.ll
index 471b04e6f3..f2b3e2c17d 100644
--- a/test/CodeGen/R600/fmul.ll
+++ b/test/CodeGen/R600/fmul.ll
@@ -1,23 +1,27 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; 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
 
-; CHECK: @fmul_f32
-; CHECK: MUL_IEEE * {{T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
-define void @fmul_f32() {
-   %r0 = call float @llvm.R600.load.input(i32 0)
-   %r1 = call float @llvm.R600.load.input(i32 1)
-   %r2 = fmul float %r0, %r1
-   call void @llvm.AMDGPU.store.output(float %r2, i32 0)
-   ret void
+; R600-CHECK: @fmul_f32
+; R600-CHECK: MUL_IEEE * {{T[0-9]+\.[XYZW]}}, KC0[2].Z, KC0[2].W
+; SI-CHECK: @fmul_f32
+; SI-CHECK: V_MUL_F32
+define void @fmul_f32(float addrspace(1)* %out, float %a, float %b) {
+entry:
+  %0 = fmul float %a, %b
+  store float %0, float addrspace(1)* %out
+  ret void
 }
 
 declare float @llvm.R600.load.input(i32) readnone
 
 declare void @llvm.AMDGPU.store.output(float, i32)
 
-; CHECK: @fmul_v2f32
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+; R600-CHECK: @fmul_v2f32
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+; SI-CHECK: @fmul_v2f32
+; SI-CHECK: V_MUL_F32
+; SI-CHECK: V_MUL_F32
 define void @fmul_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
 entry:
   %0 = fmul <2 x float> %a, %b
@@ -25,12 +29,16 @@ entry:
   ret void
 }
 
-; CHECK: @fmul_v4f32
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-
+; R600-CHECK: @fmul_v4f32
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; R600-CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; SI-CHECK: @fmul_v4f32
+; SI-CHECK: V_MUL_F32
+; SI-CHECK: V_MUL_F32
+; SI-CHECK: V_MUL_F32
+; SI-CHECK: V_MUL_F32
 define void @fmul_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
   %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
   %a = load <4 x float> addrspace(1) * %in
diff --git a/test/CodeGen/R600/fsub.ll b/test/CodeGen/R600/fsub.ll
index b45aafff5a..1608c3aced 100644
--- a/test/CodeGen/R600/fsub.ll
+++ b/test/CodeGen/R600/fsub.ll
@@ -1,23 +1,27 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; 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
 
-; CHECK: @fsub_f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
-
-define void @fsub_f32() {
-   %r0 = call float @llvm.R600.load.input(i32 0)
-   %r1 = call float @llvm.R600.load.input(i32 1)
-   %r2 = fsub float %r0, %r1
-   call void @llvm.AMDGPU.store.output(float %r2, i32 0)
-   ret void
+; R600-CHECK: @fsub_f32
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW]}}, KC0[2].Z, -KC0[2].W
+; SI-CHECK: @fsub_f32
+; SI-CHECK: V_SUB_F32
+define void @fsub_f32(float addrspace(1)* %out, float %a, float %b) {
+entry:
+  %0 = fsub float %a, %b
+  store float %0, float addrspace(1)* %out
+  ret void
 }
 
 declare float @llvm.R600.load.input(i32) readnone
 
 declare void @llvm.AMDGPU.store.output(float, i32)
 
-; CHECK: @fsub_v2f32
-; CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[3].X, -KC0[3].Z
-; CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[2].W, -KC0[3].Y
+; R600-CHECK: @fsub_v2f32
+; R600-CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[3].X, -KC0[3].Z
+; R600-CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[2].W, -KC0[3].Y
+; SI-CHECK: @fsub_v2f32
+; SI-CHECK: V_SUB_F32
+; SI-CHECK: V_SUB_F32
 define void @fsub_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
 entry:
   %0 = fsub <2 x float> %a, %b
@@ -25,11 +29,16 @@ entry:
   ret void
 }
 
-; CHECK: @fsub_v4f32
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; R600-CHECK: @fsub_v4f32
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; R600-CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; SI-CHECK: @fsub_v4f32
+; SI-CHECK: V_SUB_F32
+; SI-CHECK: V_SUB_F32
+; SI-CHECK: V_SUB_F32
+; SI-CHECK: V_SUB_F32
 define void @fsub_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
   %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
   %a = load <4 x float> addrspace(1) * %in