[llvm] r220062 - R600/SI: Allow comuting fp immediates

Matt Arsenault Matthew.Arsenault at amd.com
Fri Oct 17 11:00:40 PDT 2014 

Author: arsenm
Date: Fri Oct 17 13:00:39 2014
New Revision: 220062

URL: http://llvm.org/viewvc/llvm-project?rev=220062&view=rev
Log:
R600/SI: Allow comuting fp immediates

Modified:
    llvm/trunk/lib/Target/R600/SIInstrInfo.cpp
    llvm/trunk/test/CodeGen/R600/imm.ll

Modified: llvm/trunk/lib/Target/R600/SIInstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/SIInstrInfo.cpp?rev=220062&r1=220061&r2=220062&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/SIInstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/R600/SIInstrInfo.cpp Fri Oct 17 13:00:39 2014
@@ -712,8 +712,8 @@ MachineInstr *SIInstrInfo::commuteInstru
     return nullptr;
 
   if (!Src1.isReg()) {
-    // XXX: Commute instructions with FPImm operands
-    if (NewMI || !Src1.isImm() ||
+    // Allow commuting instructions with Imm or FPImm operands.
+    if (NewMI || (!Src1.isImm() && !Src1.isFPImm()) ||
        (!isVOP2(MI->getOpcode()) && !isVOP3(MI->getOpcode()))) {
       return nullptr;
     }
@@ -733,7 +733,13 @@ MachineInstr *SIInstrInfo::commuteInstru
 
     unsigned Reg = Src0.getReg();
     unsigned SubReg = Src0.getSubReg();
-    Src0.ChangeToImmediate(Src1.getImm());
+    if (Src1.isImm())
+      Src0.ChangeToImmediate(Src1.getImm());
+    else if (Src1.isFPImm())
+      Src0.ChangeToFPImmediate(Src1.getFPImm());
+    else
+      llvm_unreachable("Should only have immediates");
+
     Src1.ChangeToRegister(Reg, false);
     Src1.setSubReg(SubReg);
   } else {

Modified: llvm/trunk/test/CodeGen/R600/imm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/imm.ll?rev=220062&r1=220061&r2=220062&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/R600/imm.ll (original)
+++ llvm/trunk/test/CodeGen/R600/imm.ll Fri Oct 17 13:00:39 2014
@@ -104,7 +104,7 @@ define void @store_literal_imm_f32(float
 
 ; CHECK-LABEL: {{^}}add_inline_imm_0.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], 0.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], 0.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_0.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, 0.0
@@ -114,7 +114,7 @@ define void @add_inline_imm_0.0_f32(floa
 
 ; CHECK-LABEL: {{^}}add_inline_imm_0.5_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], 0.5{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], 0.5, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_0.5_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, 0.5
@@ -124,7 +124,7 @@ define void @add_inline_imm_0.5_f32(floa
 
 ; CHECK-LABEL: {{^}}add_inline_imm_neg_0.5_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], -0.5{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], -0.5, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_neg_0.5_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, -0.5
@@ -134,7 +134,7 @@ define void @add_inline_imm_neg_0.5_f32(
 
 ; CHECK-LABEL: {{^}}add_inline_imm_1.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], 1.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], 1.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_1.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, 1.0
@@ -144,7 +144,7 @@ define void @add_inline_imm_1.0_f32(floa
 
 ; CHECK-LABEL: {{^}}add_inline_imm_neg_1.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], -1.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], -1.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_neg_1.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, -1.0
@@ -154,7 +154,7 @@ define void @add_inline_imm_neg_1.0_f32(
 
 ; CHECK-LABEL: {{^}}add_inline_imm_2.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], 2.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], 2.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_2.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, 2.0
@@ -164,7 +164,7 @@ define void @add_inline_imm_2.0_f32(floa
 
 ; CHECK-LABEL: {{^}}add_inline_imm_neg_2.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], -2.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], -2.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_neg_2.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, -2.0
@@ -174,7 +174,7 @@ define void @add_inline_imm_neg_2.0_f32(
 
 ; CHECK-LABEL: {{^}}add_inline_imm_4.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], 4.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], 4.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_4.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, 4.0
@@ -184,10 +184,32 @@ define void @add_inline_imm_4.0_f32(floa
 
 ; CHECK-LABEL: {{^}}add_inline_imm_neg_4.0_f32
 ; CHECK: S_LOAD_DWORD [[VAL:s[0-9]+]]
-; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], [[VAL]], -4.0{{$}}
+; CHECK: V_ADD_F32_e64 [[REG:v[0-9]+]], -4.0, [[VAL]]{{$}}
 ; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
 define void @add_inline_imm_neg_4.0_f32(float addrspace(1)* %out, float %x) {
   %y = fadd float %x, -4.0
   store float %y, float addrspace(1)* %out
   ret void
 }
+
+; CHECK-LABEL: @commute_add_inline_imm_0.5_f32
+; CHECK: BUFFER_LOAD_DWORD [[VAL:v[0-9]+]]
+; CHECK: V_ADD_F32_e32 [[REG:v[0-9]+]], 0.5, [[VAL]]
+; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
+define void @commute_add_inline_imm_0.5_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
+  %x = load float addrspace(1)* %in
+  %y = fadd float %x, 0.5
+  store float %y, float addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: @commute_add_literal_f32
+; CHECK: BUFFER_LOAD_DWORD [[VAL:v[0-9]+]]
+; CHECK: V_ADD_F32_e32 [[REG:v[0-9]+]], 0x44800000, [[VAL]]
+; CHECK-NEXT: BUFFER_STORE_DWORD [[REG]]
+define void @commute_add_literal_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
+  %x = load float addrspace(1)* %in
+  %y = fadd float %x, 1024.0
+  store float %y, float addrspace(1)* %out
+  ret void
+}

More information about the llvm-commits mailing list