AMDGPU/GlobalISel: Fix some legalization of < dword vector stores

This avoids many instances of failing to legalize a vector truncstore
of <4 x s8> to 2 bytes. We don't perfectly handle every truncstore
yet, largely because the given set of legalization actions can't
actually differentiate between changing the result type and changing
the memory type.

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index c7d13c6..a00bd88 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -121,10 +121,10 @@ static LegalizeMutation bitcastToRegisterType(unsigned TypeIdx) {
     unsigned Size = Ty.getSizeInBits();
 
     LLT CoercedTy;
-    if (Size < 32) {
+    if (Size <= 32) {
       // <2 x s8> -> s16
-      assert(Size == 16);
-      CoercedTy = LLT::scalar(16);
+      // <4 x s8> -> s32
+      CoercedTy = LLT::scalar(Size);
     } else
       CoercedTy = LLT::scalarOrVector(Size / 32, 32);
 
@@ -982,15 +982,14 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
     Actions.bitcastIf(
       [=](const LegalityQuery &Query) -> bool {
         const LLT Ty = Query.Types[0];
+        const unsigned Size = Ty.getSizeInBits();
 
-        // Do not cast an extload/truncstore.
-        if (Ty.getSizeInBits() != Query.MMODescrs[0].SizeInBits)
-          return false;
+        if (Size != Query.MMODescrs[0].SizeInBits)
+          return Size <= 32 && Ty.isVector();
 
         if (loadStoreBitcastWorkaround(Ty) && isRegisterType(Ty))
           return true;
-        const unsigned Size = Ty.getSizeInBits();
-        return Ty.isVector() && isRegisterSize(Size) &&
+        return Ty.isVector() && (Size <= 32 || isRegisterSize(Size)) &&
                !isRegisterVectorElementType(Ty.getElementType());
       }, bitcastToRegisterType(0));
 

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-constant.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-constant.mir
index b85f3ad..3233c27 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-constant.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-constant.mir
@@ -4008,43 +4008,102 @@ body: |
 
     ; CI-LABEL: name: test_load_constant_v2s8_align4
     ; CI: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_constant_v2s8_align4
     ; VI: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_constant_v2s8_align4
     ; GFX9: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_constant_v2s8_align4
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_constant_v2s8_align4
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, align 4, addrspace 4)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p4) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 4, addrspace 4)
@@ -4061,43 +4120,102 @@ body: |
 
     ; CI-LABEL: name: test_load_constant_v2s8_align2
     ; CI: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_constant_v2s8_align2
     ; VI: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_constant_v2s8_align2
     ; GFX9: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_constant_v2s8_align2
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_constant_v2s8_align2
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p4) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p4) :: (load 2, addrspace 4)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p4) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 4)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-flat.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-flat.mir
index aaa665e..9dbbede 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-flat.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-flat.mir
@@ -4028,43 +4028,102 @@ body: |
 
     ; CI-LABEL: name: test_load_flat_v2s8_align4
     ; CI: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_flat_v2s8_align4
     ; VI: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_flat_v2s8_align4
     ; GFX9: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_flat_v2s8_align4
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_flat_v2s8_align4
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2, align 4)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p0) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 4, addrspace 0)
@@ -4081,43 +4140,102 @@ body: |
 
     ; CI-LABEL: name: test_load_flat_v2s8_align2
     ; CI: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_flat_v2s8_align2
     ; VI: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_flat_v2s8_align2
     ; GFX9: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_flat_v2s8_align2
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_flat_v2s8_align2
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p0) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p0) :: (load 2)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p0) :: (load 2)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p0) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 0)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-global.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-global.mir
index d16e963..275a4ec 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-global.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-global.mir
@@ -3745,51 +3745,123 @@ body: |
 
     ; SI-LABEL: name: test_load_global_v2s8_align4
     ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; SI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; SI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; SI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; SI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-HSA-LABEL: name: test_load_global_v2s8_align4
     ; CI-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; CI-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; CI-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; CI-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-HSA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-HSA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-HSA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-HSA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-HSA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_global_v2s8_align4
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_global_v2s8_align4
     ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-HSA-LABEL: name: test_load_global_v2s8_align4
     ; GFX9-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; GFX9-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; GFX9-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-HSA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-HSA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-HSA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_global_v2s8_align4
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 4, addrspace 1)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p1) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 4, addrspace 1)
@@ -3806,51 +3878,123 @@ body: |
 
     ; SI-LABEL: name: test_load_global_v2s8_align2
     ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; SI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; SI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; SI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; SI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-HSA-LABEL: name: test_load_global_v2s8_align2
     ; CI-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; CI-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; CI-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; CI-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-HSA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-HSA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-HSA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-HSA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-HSA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_global_v2s8_align2
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; CI-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; CI-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; CI-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-MESA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-MESA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-MESA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-MESA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-MESA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_global_v2s8_align2
     ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-HSA-LABEL: name: test_load_global_v2s8_align2
     ; GFX9-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; GFX9-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; GFX9-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-HSA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-HSA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-HSA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_global_v2s8_align2
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
-    ; GFX9-MESA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-MESA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-MESA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-MESA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, addrspace 1)
+    ; GFX9-MESA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-MESA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-MESA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-MESA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-MESA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-MESA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-MESA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-MESA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-MESA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-MESA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-MESA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-MESA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-MESA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-MESA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-MESA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-MESA: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p1) = COPY $vgpr0_vgpr1
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 1)
@@ -3885,11 +4029,24 @@ body: |
     ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-HSA-LABEL: name: test_load_global_v2s8_align1
     ; CI-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; CI-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 1, addrspace 1)
-    ; CI-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 1, addrspace 1)
+    ; CI-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-HSA: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-HSA: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-HSA: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-HSA: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-HSA: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-MESA-LABEL: name: test_load_global_v2s8_align1
     ; CI-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
@@ -3927,11 +4084,22 @@ body: |
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-HSA-LABEL: name: test_load_global_v2s8_align1
     ; GFX9-HSA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p1) :: (load 2, align 1, addrspace 1)
-    ; GFX9-HSA: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9-HSA: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9-HSA: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9-HSA: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p1) :: (load 2, align 1, addrspace 1)
+    ; GFX9-HSA: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9-HSA: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9-HSA: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9-HSA: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9-HSA: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9-HSA: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9-HSA: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9-HSA: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9-HSA: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9-HSA: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9-HSA: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9-HSA: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9-HSA: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9-HSA: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9-HSA: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9-HSA: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-MESA-LABEL: name: test_load_global_v2s8_align1
     ; GFX9-MESA: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-local.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-local.mir
index 001bfc0..a465c05 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-local.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-local.mir
@@ -4909,43 +4909,104 @@ body: |
 
     ; SI-LABEL: name: test_load_local_v2s8_align2
     ; SI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
-    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
-    ; SI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; SI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; SI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; SI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-LABEL: name: test_load_local_v2s8_align2
     ; CI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-DS128-LABEL: name: test_load_local_v2s8_align2
     ; CI-DS128: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
-    ; CI-DS128: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
-    ; CI-DS128: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI-DS128: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI-DS128: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI-DS128: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
+    ; CI-DS128: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI-DS128: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI-DS128: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI-DS128: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI-DS128: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI-DS128: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI-DS128: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI-DS128: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI-DS128: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI-DS128: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI-DS128: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI-DS128: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI-DS128: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI-DS128: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI-DS128: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI-DS128: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI-DS128: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI-DS128: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_local_v2s8_align2
     ; VI: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_local_v2s8_align2
     ; GFX9: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p3) :: (load 2, addrspace 3)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p3) = COPY $vgpr0
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 3)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-private.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-private.mir
index 5a598e2..9c43f7a 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-private.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-load-private.mir
@@ -4069,35 +4069,83 @@ body: |
 
     ; SI-LABEL: name: test_load_private_v2s8_align2
     ; SI: [[COPY:%[0-9]+]]:_(p5) = COPY $vgpr0
-    ; SI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
-    ; SI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; SI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; SI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; SI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; SI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; SI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; SI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; CI-LABEL: name: test_load_private_v2s8_align2
     ; CI: [[COPY:%[0-9]+]]:_(p5) = COPY $vgpr0
-    ; CI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
-    ; CI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; CI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; CI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; CI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
+    ; CI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; CI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; CI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; CI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; CI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; CI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; CI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; CI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; CI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
+    ; CI: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; CI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C4]]
+    ; CI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY1]](s32)
+    ; CI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; CI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; CI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; CI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; VI-LABEL: name: test_load_private_v2s8_align2
     ; VI: [[COPY:%[0-9]+]]:_(p5) = COPY $vgpr0
-    ; VI: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
-    ; VI: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; VI: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; VI: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; VI: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; VI: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; VI: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; VI: $vgpr0 = COPY [[ANYEXT]](s32)
     ; GFX9-LABEL: name: test_load_private_v2s8_align2
     ; GFX9: [[COPY:%[0-9]+]]:_(p5) = COPY $vgpr0
-    ; GFX9: [[LOAD:%[0-9]+]]:_(<4 x s8>) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
-    ; GFX9: [[EXTRACT:%[0-9]+]]:_(<2 x s8>) = G_EXTRACT [[LOAD]](<4 x s8>), 0
-    ; GFX9: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[EXTRACT]](<2 x s8>)
-    ; GFX9: [[MV:%[0-9]+]]:_(s16) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8)
-    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[MV]](s16)
+    ; GFX9: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[COPY]](p5) :: (load 2, addrspace 5)
+    ; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C]](s32)
+    ; GFX9: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C1]](s32)
+    ; GFX9: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[LOAD]], [[C2]](s32)
+    ; GFX9: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[LOAD]](s32)
+    ; GFX9: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C3]]
+    ; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
+    ; GFX9: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C3]]
+    ; GFX9: [[C4:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; GFX9: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C4]](s16)
+    ; GFX9: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
     ; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
     %0:_(p5) = COPY $vgpr0
     %1:_(<2 x s8>) = G_LOAD %0 :: (load 2, align 2, addrspace 5)

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-store.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-store.mir
index eceec6a..55641ca 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-store.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-store.mir
@@ -387,25 +387,685 @@ body: |
 ...
 
 ---
-name: test_store_global_v3s8_align4
+name: test_store_global_v2s8_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_store_global_v2s8_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; SI: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; SI: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: G_STORE [[COPY2]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v2s8_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; VI: G_STORE [[COPY1]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; VI: G_STORE [[COPY2]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s8>) = G_IMPLICIT_DEF
+    G_STORE %1, %0 :: (store 2, addrspace 1, align 1)
+
+...
+
+---
+name: test_store_global_v2s8_align2
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_store_global_v2s8_align2
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v2s8_align2
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s8>) = G_IMPLICIT_DEF
+    G_STORE %1, %0 :: (store 2, addrspace 1, align 2)
+
+...
+
+---
+name: test_store_global_v2s8_align4
 body: |
   bb.0:
     liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
 
+    ; SI-LABEL: name: test_store_global_v2s8_align4
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, align 4, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v2s8_align4
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, align 4, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s8>) = G_IMPLICIT_DEF
+    G_STORE %1, %0 :: (store 2, addrspace 1, align 4)
+
+...
+
+---
+name: test_store_global_v3s8_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
+
+    ; SI-LABEL: name: test_store_global_v3s8_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; SI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY2]](s32)
+    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC2]]
+    ; SI: [[COPY4:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[COPY4]](s16)
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT]], [[C1]](s32)
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY4]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; SI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: G_STORE [[COPY5]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; SI: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C4]](s64)
+    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: G_STORE [[COPY6]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v3s8_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; VI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[COPY2:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; VI: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[COPY2]], [[C1]](s16)
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR]](s16)
+    ; VI: G_STORE [[ANYEXT1]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; VI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: G_STORE [[COPY3]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    %2:_(<3 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 1)
+
+...
+
+---
+name: test_store_global_v3s8_align2
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
+
+    ; SI-LABEL: name: test_store_global_v3s8_align2
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; SI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC2]]
+    ; SI: [[COPY3:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY3]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; SI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: G_STORE [[COPY4]](s32), [[PTR_ADD]](p1) :: (store 1 + 2, align 2, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v3s8_align2
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; VI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[COPY2:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: G_STORE [[COPY3]](s32), [[PTR_ADD]](p1) :: (store 1 + 2, align 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    %2:_(<3 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 2)
+
+...
+
+---
+name: test_store_global_v3s8_align4
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
+
     ; SI-LABEL: name: test_store_global_v3s8_align4
     ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; SI: [[DEF:%[0-9]+]]:_(<3 x s8>) = G_IMPLICIT_DEF
-    ; SI: [[DEF1:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
-    ; SI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF1]], [[DEF]](<3 x s8>), 0
-    ; SI: G_STORE [[INSERT]](<4 x s8>), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; SI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; SI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
     ; VI-LABEL: name: test_store_global_v3s8_align4
     ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
-    ; VI: [[DEF:%[0-9]+]]:_(<3 x s8>) = G_IMPLICIT_DEF
-    ; VI: [[DEF1:%[0-9]+]]:_(<4 x s8>) = G_IMPLICIT_DEF
-    ; VI: [[INSERT:%[0-9]+]]:_(<4 x s8>) = G_INSERT [[DEF1]], [[DEF]](<3 x s8>), 0
-    ; VI: G_STORE [[INSERT]](<4 x s8>), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; VI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; VI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    %2:_(<3 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 4)
+
+...
+
+---
+name: test_store_global_v4s8_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_store_global_v4s8_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; SI: G_STORE [[COPY2]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; SI: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: G_STORE [[COPY3]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; SI: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C1]](s64)
+    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: G_STORE [[COPY4]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    ; SI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 3
+    ; SI: [[PTR_ADD2:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; SI: G_STORE [[COPY5]](s32), [[PTR_ADD2]](p1) :: (store 1 + 3, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v4s8_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; VI: G_STORE [[COPY2]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C]](s64)
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; VI: G_STORE [[COPY3]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; VI: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C1]](s64)
+    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: G_STORE [[COPY4]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 3
+    ; VI: [[PTR_ADD2:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; VI: G_STORE [[COPY5]](s32), [[PTR_ADD2]](p1) :: (store 1 + 3, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 4, addrspace 1, align 1)
+
+...
+
+---
+name: test_store_global_v4s8_align2
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_store_global_v4s8_align2
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY2]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV2]](s32)
+    ; SI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C2]]
+    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C1]](s32)
+    ; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SHL1]](s32)
+    ; SI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[TRUNC3]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; SI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; SI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[OR1]](s16)
+    ; SI: G_STORE [[ANYEXT1]](s32), [[PTR_ADD]](p1) :: (store 2 + 2, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v4s8_align2
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV2]](s32)
+    ; VI: [[AND2:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[UV3]](s32)
+    ; VI: [[AND3:%[0-9]+]]:_(s16) = G_AND [[TRUNC3]], [[C]]
+    ; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[AND3]], [[C1]](s16)
+    ; VI: [[OR1:%[0-9]+]]:_(s16) = G_OR [[AND2]], [[SHL1]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[OR1]](s16)
+    ; VI: G_STORE [[ANYEXT1]](s32), [[PTR_ADD]](p1) :: (store 2 + 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 4, addrspace 1, align 2)
+
+...
+
+---
+name: test_store_global_v4s8_align4
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_store_global_v4s8_align4
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
+    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C2]](s32)
+    ; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; SI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; SI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C3]](s32)
+    ; SI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; SI: G_STORE [[OR2]](s32), [[COPY]](p1) :: (store 4, addrspace 1)
+    ; VI-LABEL: name: test_store_global_v4s8_align4
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C]]
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
+    ; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY4]], [[C]]
+    ; VI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C2]](s32)
+    ; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
+    ; VI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[UV3]](s32)
+    ; VI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY5]], [[C]]
+    ; VI: [[C3:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
+    ; VI: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C3]](s32)
+    ; VI: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
+    ; VI: G_STORE [[OR2]](s32), [[COPY]](p1) :: (store 4, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 4, addrspace 1, align 4)
+
+...
+
+---
+name: test_truncstore_global_v2s8_to_1_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_truncstore_global_v2s8_to_1_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY1]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC1]]
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v2s8_to_1_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[DEF:%[0-9]+]]:_(<2 x s32>) = G_IMPLICIT_DEF
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[DEF]](<2 x s32>)
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC]], [[C]]
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s8>) = G_IMPLICIT_DEF
+    G_STORE %1, %0 :: (store 1, addrspace 1, align 1)
+
+...
+
+---
+name: test_truncstore_global_v3s8_to_1_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
+
+    ; SI-LABEL: name: test_truncstore_global_v3s8_to_1_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; SI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; SI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v3s8_to_1_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; VI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; VI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 1, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    %2:_(<3 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 1, addrspace 1, align 1)
+
+...
+
+---
+name: test_truncstore_global_v3s8_to_2_align2
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
+
+    ; SI-LABEL: name: test_truncstore_global_v3s8_to_2_align2
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; SI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; SI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; SI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v3s8_to_2_align2
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    ; VI: [[TRUNC:%[0-9]+]]:_(<3 x s8>) = G_TRUNC [[COPY1]](<3 x s32>)
+    ; VI: [[BITCAST:%[0-9]+]]:_(s24) = G_BITCAST [[TRUNC]](<3 x s8>)
+    ; VI: G_STORE [[BITCAST]](s24), [[COPY]](p1) :: (store 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
+    %2:_(<3 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 2, addrspace 1, align 2)
+
+...
+
+
+---
+name: test_truncstore_global_v4s8_to_3_align1
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_truncstore_global_v4s8_to_3_align1
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; SI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C1]](s32)
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY3]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[COPY2]](s32)
+    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC2]]
+    ; SI: [[COPY4:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; SI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[COPY4]](s16)
+    ; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[ZEXT]], [[C1]](s32)
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY4]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; SI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
+    ; SI: G_STORE [[COPY5]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; SI: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C4]](s64)
+    ; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: G_STORE [[COPY6]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v4s8_to_3_align1
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; VI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[COPY2:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; VI: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[COPY2]], [[C1]](s16)
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 1, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR]](s16)
+    ; VI: G_STORE [[ANYEXT1]](s32), [[PTR_ADD]](p1) :: (store 1 + 1, addrspace 1)
+    ; VI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD1:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: G_STORE [[COPY3]](s32), [[PTR_ADD1]](p1) :: (store 1 + 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 1)
+
+...
+
+---
+name: test_truncstore_global_v4s8_to_3_align2
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_truncstore_global_v4s8_to_3_align2
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; SI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; SI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; SI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; SI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
+    ; SI: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[UV1]](s32)
+    ; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY2]], [[C2]]
+    ; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C1]](s32)
+    ; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SHL]](s32)
+    ; SI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[TRUNC2]]
+    ; SI: [[COPY3:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY3]](s16)
+    ; SI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; SI: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; SI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C3]](s64)
+    ; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; SI: G_STORE [[COPY4]](s32), [[PTR_ADD]](p1) :: (store 1 + 2, align 2, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v4s8_to_3_align2
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
+    ; VI: [[TRUNC:%[0-9]+]]:_(s8) = G_TRUNC [[UV2]](s32)
+    ; VI: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+    ; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
+    ; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[UV]](s32)
+    ; VI: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC1]], [[C]]
+    ; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[UV1]](s32)
+    ; VI: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC2]], [[C]]
+    ; VI: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
+    ; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C1]](s16)
+    ; VI: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
+    ; VI: [[COPY2:%[0-9]+]]:_(s16) = COPY [[OR]](s16)
+    ; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY2]](s16)
+    ; VI: G_STORE [[ANYEXT]](s32), [[COPY]](p1) :: (store 2, addrspace 1)
+    ; VI: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; VI: [[PTR_ADD:%[0-9]+]]:_(p1) = G_PTR_ADD [[COPY]], [[C2]](s64)
+    ; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[UV2]](s32)
+    ; VI: G_STORE [[COPY3]](s32), [[PTR_ADD]](p1) :: (store 1 + 2, align 2, addrspace 1)
+    %0:_(p1) = COPY $vgpr0_vgpr1
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 2)
+
+...
+
+---
+name: test_truncstore_global_v4s8_to_3_align4
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+    ; SI-LABEL: name: test_truncstore_global_v4s8_to_3_align4
+    ; SI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; SI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; SI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[COPY1]](<4 x s32>)
+    ; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[TRUNC]](<4 x s8>)
+    ; SI: G_STORE [[BITCAST]](s32), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
+    ; VI-LABEL: name: test_truncstore_global_v4s8_to_3_align4
+    ; VI: [[COPY:%[0-9]+]]:_(p1) = COPY $vgpr0_vgpr1
+    ; VI: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    ; VI: [[TRUNC:%[0-9]+]]:_(<4 x s8>) = G_TRUNC [[COPY1]](<4 x s32>)
+    ; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[TRUNC]](<4 x s8>)
+    ; VI: G_STORE [[BITCAST]](s32), [[COPY]](p1) :: (store 3, align 4, addrspace 1)
     %0:_(p1) = COPY $vgpr0_vgpr1
-    %1:_(<3 x s8>) = G_IMPLICIT_DEF
-    G_STORE %1, %0 :: (store 3, addrspace 1, align 4)
+    %1:_(<4 x s32>) = COPY $vgpr2_vgpr3_vgpr4_vgpr5
+    %2:_(<4 x s8>) = G_TRUNC %1
+    G_STORE %2, %0 :: (store 3, addrspace 1, align 4)
 
 ...