From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 5 Apr 2019 14:03:04 +0000 (+0000)
Subject: AMDGPU/GlobalISel: Fix non-power-of-2 select
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=6c7dd5967af1a9cb91f5a4e4c11441249b96c241;p=llvm

AMDGPU/GlobalISel: Fix non-power-of-2 select

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@357762 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 61d38ba0e6f..5835ccec579 100644
--- a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -486,6 +486,7 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
     .clampMaxNumElements(0, LocalPtr, 2)
     .clampMaxNumElements(0, PrivatePtr, 2)
     .scalarize(0)
+    .widenScalarToNextPow2(0)
     .legalIf(all(isPointer(0), typeIs(1, S1)));
 
   // TODO: Only the low 4/5/6 bits of the shift amount are observed, so we can
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-select.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-select.mir
index 8d4dfc67b68..cb37a22e832 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-select.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-select.mir
@@ -49,6 +49,34 @@ body: |
 
 ...
 
+---
+name: test_select_s48
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: test_select_s48
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(ne), [[C]](s32), [[COPY]]
+    ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY [[C1]](s64)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY [[C2]](s64)
+    ; CHECK: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY1]], [[COPY2]]
+    ; CHECK: [[COPY3:%[0-9]+]]:_(s64) = COPY [[SELECT]](s64)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[COPY3]](s64)
+    %0:_(s32) = G_CONSTANT i32 0
+    %1:_(s32) = COPY $vgpr0
+
+    %2:_(s1) = G_ICMP intpred(ne), %0, %1
+    %3:_(s48) = G_CONSTANT i48 1
+    %4:_(s48) = G_CONSTANT i48 2
+    %5:_(s48) = G_SELECT %2, %3, %4
+    %6:_(s64) = G_ANYEXT %5
+    $vgpr0_vgpr1 = COPY %6
+
+...
+
 ---
 name: test_select_s16
 body: |