[AArch64][GlobalISel] Promote scalar G_SHL constant shift amounts to …

…s64. This was supposed to be done in the first place as is currently the case for G_ASHR and G_LSHR but was forgotten when the original shift legalization overhaul was done last year. This was exposed because we started falling back on s32 = s32, s64 SHLs due to a recent combiner change. Gives a very minor (0.1%) code size -O0 improvement on consumer-typeset.
llvm · Dec 18, 2020 · 43ff75f · 43ff75f
1 parent 9caca72
commit 43ff75f
Show file tree

Hide file tree

Showing 5 changed files with 28 additions and 20 deletions.
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -97,15 +97,25 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .moreElementsToNextPow2(0);
 
   getActionDefinitionsBuilder(G_SHL)
-    .legalFor({{s32, s32}, {s64, s64},
-               {v2s32, v2s32}, {v4s32, v4s32}, {v2s64, v2s64}})
-    .clampScalar(1, s32, s64)
-    .clampScalar(0, s32, s64)
-    .widenScalarToNextPow2(0)
-    .clampNumElements(0, v2s32, v4s32)
-    .clampNumElements(0, v2s64, v2s64)
-    .moreElementsToNextPow2(0)
-    .minScalarSameAs(1, 0);
+      .customIf([=](const LegalityQuery &Query) {
+        const auto &SrcTy = Query.Types[0];
+        const auto &AmtTy = Query.Types[1];
+        return !SrcTy.isVector() && SrcTy.getSizeInBits() == 32 &&
+               AmtTy.getSizeInBits() == 32;
+      })
+      .legalFor({{s32, s32},
+                 {s64, s64},
+                 {s32, s64},
+                 {v2s32, v2s32},
+                 {v4s32, v4s32},
+                 {v2s64, v2s64}})
+      .clampScalar(1, s32, s64)
+      .clampScalar(0, s32, s64)
+      .widenScalarToNextPow2(0)
+      .clampNumElements(0, v2s32, v4s32)
+      .clampNumElements(0, v2s64, v2s64)
+      .moreElementsToNextPow2(0)
+      .minScalarSameAs(1, 0);
 
   getActionDefinitionsBuilder(G_PTR_ADD)
       .legalFor({{p0, s64}, {v2p0, v2s64}})

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-merge-values.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-merge-values.mir
@@ -6,11 +6,12 @@ name:            test_merge_s4
 body: |
   bb.0:
     ; CHECK-LABEL: name: test_merge_s4
-    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 4
+    ; CHECK: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 4
     ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 15
     ; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
     ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[C2]], [[C1]]
-    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C]](s32)
+    ; CHECK: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 4
+    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C3]](s64)
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY [[C2]](s32)
     ; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C1]]
     ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[SHL]](s32)

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-non-pow2-load-store.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-non-pow2-load-store.mir
@@ -28,12 +28,11 @@ body:             |
     ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
     ; CHECK: [[PTR_ADD:%[0-9]+]]:_(p0) = G_PTR_ADD [[COPY]], [[C1]](s64)
     ; CHECK: [[LOAD:%[0-9]+]]:_(s32) = G_LOAD [[PTR_ADD]](p0) :: (load 1 from %ir.ptr + 2, align 4)
-    ; CHECK: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
-    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LOAD]], [[C2]](s32)
+    ; CHECK: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
+    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[LOAD]], [[C2]](s64)
     ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[SHL]], [[ZEXTLOAD]]
     ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[OR]](s32)
-    ; CHECK: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
-    ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY2]], [[C3]](s64)
+    ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY2]], [[C2]](s64)
     ; CHECK: [[PTR_ADD1:%[0-9]+]]:_(p0) = G_PTR_ADD [[COPY1]], [[C1]](s64)
     ; CHECK: G_STORE [[COPY2]](s32), [[COPY1]](p0) :: (store 2 into %ir.ptr2, align 4)
     ; CHECK: G_STORE [[LSHR]](s32), [[PTR_ADD1]](p0) :: (store 1 into %ir.ptr2 + 2, align 4)

diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-shift.mir
@@ -235,8 +235,8 @@ body:             |
 
     ; CHECK-LABEL: name: shl_cimm_32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
-    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
-    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
+    ; CHECK: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[C]](s64)
     ; CHECK: $w0 = COPY [[SHL]](s32)
     ; CHECK: RET_ReallyLR implicit $w0
     %0:_(s32) = COPY $w0

diff --git a/llvm/test/CodeGen/AArch64/arm64-clrsb.ll b/llvm/test/CodeGen/AArch64/arm64-clrsb.ll
@@ -21,10 +21,8 @@ entry:
 ; CHECK-LABEL: clrsb32
 ; CHECK:   cls [[TEMP:w[0-9]+]], [[TEMP]]
 
-; FIXME: We should produce the same result here to save some code size. After
-; that, we can remove the GISEL special casing.
 ; GISEL-LABEL: clrsb32
-; GISEL: clz
+; GISEL: cls [[TEMP:w[0-9]+]], [[TEMP]]
 }
 
 ; Function Attrs: nounwind ssp