R600/SI: Get rid of FCLAMP_SI pseudo

It's not necessary. Also use complex patterns to allow
src modifier usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@221916 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/R600/AMDGPUISelDAGToDAG.cpp

@@ -111,6 +111,9 @@ private:
   bool SelectVOP3Mods0(SDValue In, SDValue &Src, SDValue &SrcMods,
                        SDValue &Clamp, SDValue &Omod) const;
 
+  bool SelectVOP3Mods0Clamp(SDValue In, SDValue &Src, SDValue &SrcMods,
+                            SDValue &Omod) const;
+
   SDNode *SelectADD_SUB_I64(SDNode *N);
   SDNode *SelectDIV_SCALE(SDNode *N);
 
@@ -1129,6 +1132,15 @@ bool AMDGPUDAGToDAGISel::SelectVOP3Mods0(SDValue In, SDValue &Src,
   return SelectVOP3Mods(In, Src, SrcMods);
 }
 
+bool AMDGPUDAGToDAGISel::SelectVOP3Mods0Clamp(SDValue In, SDValue &Src,
+                                              SDValue &SrcMods,
+                                              SDValue &Omod) const {
+  // FIXME: Handle Omod
+  Omod = CurDAG->getTargetConstant(0, MVT::i32);
+
+  return SelectVOP3Mods(In, Src, SrcMods);
+}
+
 void AMDGPUDAGToDAGISel::PostprocessISelDAG() {
   const AMDGPUTargetLowering& Lowering =
     *static_cast<const AMDGPUTargetLowering*>(getTargetLowering());

lib/Target/R600/SIISelLowering.cpp

@@ -635,20 +635,6 @@ MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
     MI->eraseFromParent();
     break;
   }
-  case AMDGPU::FCLAMP_SI: {
-    const SIInstrInfo *TII = static_cast<const SIInstrInfo *>(
-        getTargetMachine().getSubtargetImpl()->getInstrInfo());
-    DebugLoc DL = MI->getDebugLoc();
-    unsigned DestReg = MI->getOperand(0).getReg();
-    BuildMI(*BB, I, DL, TII->get(AMDGPU::V_ADD_F32_e64), DestReg)
-      .addImm(0) // SRC0 modifiers
-      .addOperand(MI->getOperand(1))
-      .addImm(0) // SRC1 modifiers
-      .addImm(0) // SRC1
-      .addImm(1) // CLAMP
-      .addImm(0); // OMOD
-    MI->eraseFromParent();
-  }
   }
   return BB;
 }

lib/Target/R600/SIInstrInfo.td

@@ -245,6 +245,7 @@ def MUBUFOffset : ComplexPattern<i64, 6, "SelectMUBUFOffset">;
 def MUBUFOffsetAtomic : ComplexPattern<i64, 4, "SelectMUBUFOffset">;
 
 def VOP3Mods0 : ComplexPattern<untyped, 4, "SelectVOP3Mods0">;
+def VOP3Mods0Clamp : ComplexPattern<untyped, 3, "SelectVOP3Mods0Clamp">;
 def VOP3Mods  : ComplexPattern<untyped, 2, "SelectVOP3Mods">;
 
 //===----------------------------------------------------------------------===//

lib/Target/R600/SIInstructions.td

@@ -2387,18 +2387,10 @@ def : BitConvert <v16f32, v16i32, VReg_512>;
 /********** Src & Dst modifiers **********/
 /********** =================== **********/
 
-def FCLAMP_SI : AMDGPUShaderInst <
-  (outs VReg_32:$dst),
-  (ins VSrc_32:$src0),
-  "fclamp_si $dst, $src0",
-  []
-> {
-  let usesCustomInserter = 1;
-}
-
 def : Pat <
-  (AMDGPUclamp f32:$src, (f32 FP_ZERO), (f32 FP_ONE)),
+  (AMDGPUclamp (VOP3Mods0Clamp f32:$src0, i32:$src0_modifiers, i32:$omod),
-  (FCLAMP_SI f32:$src)
+               (f32 FP_ZERO), (f32 FP_ONE)),
+  (V_ADD_F32_e64 $src0_modifiers, $src0, 0, 0, 1, $omod)
 >;
 
 /********** ================================ **********/

test/CodeGen/R600/llvm.AMDGPU.clamp.ll

@@ -1,6 +1,7 @@
 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -strict-whitespace -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 
+declare float @llvm.fabs.f32(float) nounwind readnone
 declare float @llvm.AMDGPU.clamp.f32(float, float, float) nounwind readnone
 declare float @llvm.AMDIL.clamp.f32(float, float, float) nounwind readnone
 
@@ -17,6 +18,43 @@ define void @clamp_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
   ret void
 }
 
+; FUNC-LABEL: {{^}}clamp_fabs_0_1_f32:
+; SI: s_load_dword [[ARG:s[0-9]+]],
+; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, |[[ARG]]| clamp{{$}}
+; SI: buffer_store_dword [[RESULT]]
+; SI: s_endpgm
+define void @clamp_fabs_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
+  %src.fabs = call float @llvm.fabs.f32(float %src) nounwind readnone
+  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fabs, float 0.0, float 1.0) nounwind readnone
+  store float %clamp, float addrspace(1)* %out, align 4
+  ret void
+}
+
+; FUNC-LABEL: {{^}}clamp_fneg_0_1_f32:
+; SI: s_load_dword [[ARG:s[0-9]+]],
+; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, -[[ARG]] clamp{{$}}
+; SI: buffer_store_dword [[RESULT]]
+; SI: s_endpgm
+define void @clamp_fneg_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
+  %src.fneg = fsub float -0.0, %src
+  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fneg, float 0.0, float 1.0) nounwind readnone
+  store float %clamp, float addrspace(1)* %out, align 4
+  ret void
+}
+
+; FUNC-LABEL: {{^}}clamp_fneg_fabs_0_1_f32:
+; SI: s_load_dword [[ARG:s[0-9]+]],
+; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, -|[[ARG]]| clamp{{$}}
+; SI: buffer_store_dword [[RESULT]]
+; SI: s_endpgm
+define void @clamp_fneg_fabs_0_1_f32(float addrspace(1)* %out, float %src) nounwind {
+  %src.fabs = call float @llvm.fabs.f32(float %src) nounwind readnone
+  %src.fneg.fabs = fsub float -0.0, %src.fabs
+  %clamp = call float @llvm.AMDGPU.clamp.f32(float %src.fneg.fabs, float 0.0, float 1.0) nounwind readnone
+  store float %clamp, float addrspace(1)* %out, align 4
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}clamp_0_1_amdil_legacy_f32:
 ; SI: s_load_dword [[ARG:s[0-9]+]],
 ; SI: v_add_f32_e64 [[RESULT:v[0-9]+]], 0, [[ARG]] clamp{{$}}