[PATCH] D45883: AMDGPU/GlobalISel: Implement select() for 32-bit G_FPTOUI

Tom Stellard via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Apr 23 20:43:02 PDT 2018


tstellar updated this revision to Diff 143676.
tstellar added a comment.

Rewrote the pactch to use the TableGen'd instruction selector.


Repository:
  rL LLVM

https://reviews.llvm.org/D45883

Files:
  lib/Target/AMDGPU/AMDGPUGISel.td
  lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
  lib/Target/AMDGPU/AMDGPUInstructionSelector.h
  test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir


Index: test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir
===================================================================
--- /dev/null
+++ test/CodeGen/AMDGPU/GlobalISel/inst-select-fptoui.mir
@@ -0,0 +1,36 @@
+# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s -check-prefixes=GCN
+
+--- |
+  define amdgpu_kernel void @fptoui(i32 addrspace(1)* %global0) {ret void}
+...
+---
+
+name:            fptoui
+legalized:       true
+regBankSelected: true
+
+# GCN-LABEL: name: fptoui
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0, $vgpr3_vgpr4
+
+    ; GCN: [[SGPR:%[0-9]+]]:sreg_32 = COPY $sgpr0
+    %0:sgpr(s32) = COPY $sgpr0
+
+    ; GCN: [[VGPR:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    %1:vgpr(s32) = COPY $vgpr0
+
+    %2:vgpr(s64) = COPY $vgpr3_vgpr4
+
+    ; fptoui s
+    ; GCN: V_CVT_U32_F32_e64 0, [[SGPR]], 0, 0
+    %3:vgpr(s32) = G_FPTOUI %0
+
+    ; fptoui v
+    ; GCN: V_CVT_U32_F32_e64 0, [[VGPR]], 0, 0
+    %4:vgpr(s32) = G_FPTOUI %1
+
+    G_STORE %3, %2 :: (store 4 into %ir.global0)
+    G_STORE %4, %2 :: (store 4 into %ir.global0)
+...
+---
Index: lib/Target/AMDGPU/AMDGPUInstructionSelector.h
===================================================================
--- lib/Target/AMDGPU/AMDGPUInstructionSelector.h
+++ lib/Target/AMDGPU/AMDGPUInstructionSelector.h
@@ -72,6 +72,9 @@
   bool selectG_STORE(MachineInstr &I) const;
   bool selectImpl(MachineInstr &I) const;
 
+  InstructionSelector::ComplexRendererFns
+  selectVOP3Mods0(MachineOperand &Root) const;
+
   const SIInstrInfo &TII;
   const SIRegisterInfo &TRI;
   const AMDGPURegisterBankInfo &RBI;
Index: lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -532,6 +532,7 @@
   switch (I.getOpcode()) {
   default:
     break;
+  case TargetOpcode::G_FPTOUI:
   case TargetOpcode::G_OR:
     return selectImpl(I, CoverageInfo);
   case TargetOpcode::G_ADD:
@@ -551,3 +552,13 @@
   }
   return false;
 }
+
+InstructionSelector::ComplexRendererFns
+AMDGPUInstructionSelector::selectVOP3Mods0(MachineOperand &Root) const {
+  return {{
+      [=](MachineInstrBuilder &MIB) { MIB.add(Root); },
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // src0_mods
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // clamp
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }  // omod
+  }};
+}
Index: lib/Target/AMDGPU/AMDGPUGISel.td
===================================================================
--- lib/Target/AMDGPU/AMDGPUGISel.td
+++ lib/Target/AMDGPU/AMDGPUGISel.td
@@ -32,3 +32,7 @@
 }
 
 defm : GISelVOP2DefaultPat <or, V_OR_B32_e64, i32>;
+
+def gi_vop3mods0 :
+    GIComplexOperandMatcher<s32, "selectVOP3Mods0">,
+    GIComplexPatternEquiv<VOP3Mods0>;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D45883.143676.patch
Type: text/x-patch
Size: 2891 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180424/f6cb8986/attachment.bin>


More information about the llvm-commits mailing list