[llvm] r353522 - AMDGPU/GlobalISel: Fix non-power-of-2 implicit_def

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Fri Feb  8 06:46:27 2019
New Revision: 353522

URL: http://llvm.org/viewvc/llvm-project?rev=353522&view=rev
Log:
AMDGPU/GlobalISel: Fix non-power-of-2 implicit_def

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
    llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-implicit-def.mir

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp?rev=353522&r1=353521&r2=353522&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp Fri Feb  8 06:46:27 2019
@@ -139,7 +139,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo
                ConstantPtr, LocalPtr, FlatPtr, PrivatePtr})
     .legalFor({LLT::vector(3, 16)})// FIXME: Hack
     .clampScalarOrElt(0, S32, S512)
-    .legalIf(isMultiple32(0));
+    .legalIf(isMultiple32(0))
+    .widenScalarToNextPow2(0, 32);
 
 
   // FIXME: i1 operands to intrinsics should always be legal, but other i1

Modified: llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-implicit-def.mir
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-implicit-def.mir?rev=353522&r1=353521&r2=353522&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-implicit-def.mir (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-implicit-def.mir Fri Feb  8 06:46:27 2019
@@ -70,6 +70,20 @@ body: |
 ...
 
 ---
+name: test_implicit_def_48
+body: |
+  bb.0:
+
+    ; CHECK-LABEL: name: test_implicit_def_48
+    ; CHECK: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
+    ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY [[DEF]](s64)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[COPY]](s64)
+    %0:_(s48) = G_IMPLICIT_DEF
+    %1:_(s64) = G_ANYEXT %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
 name: test_implicit_def_s64
 body: |
   bb.0:
@@ -82,6 +96,20 @@ body: |
 ...
 
 ---
+name: test_implicit_def_s65
+body: |
+  bb.0:
+
+    ; CHECK-LABEL: name: test_implicit_def_s65
+    ; CHECK: [[DEF:%[0-9]+]]:_(s128) = G_IMPLICIT_DEF
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s96) = G_TRUNC [[DEF]](s128)
+    ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[TRUNC]](s96)
+    %0:_(s65) = G_IMPLICIT_DEF
+    %1:_(s96) = G_ANYEXT %0
+    $vgpr0_vgpr1_vgpr2 = COPY %1
+...
+
+---
 name: test_implicit_def_s128
 body: |
   bb.0: