[X86][InstCombine] Add basic simplification support for BEXTR/BEXTRI intrinsics.

This patch adds simplification support for the BEXTR/BEXTRI intrinsics to match gcc. This only supports cases that fold to 0 or can be fully constant folded. Theoretically we could support converting to AND if the shift part is unused or to only a shift if the mask doesn't modify any bits after an equivalent shl. gcc doesn't do these transformations either.

I put this in InstCombine, but it could be done in InstSimplify. It would be the first target specific intrinsic in InstSimplify.

Differential Revision: https://reviews.llvm.org/D36063

llvm-svn: 309603

2 files changed, 229 insertions, 0 deletions

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 3046f9ba6cd..5a72373921e 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2248,6 +2248,32 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     }
     break;
 
+  case Intrinsic::x86_bmi_bextr_32:
+  case Intrinsic::x86_bmi_bextr_64:
+  case Intrinsic::x86_tbm_bextri_u32:
+  case Intrinsic::x86_tbm_bextri_u64:
+    // If the RHS is a constant we can try some simplifications.
+    if (auto *C = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
+      uint64_t Shift = C->getZExtValue();
+      uint64_t Length = (Shift >> 8) & 0xff;
+      Shift &= 0xff;
+      unsigned BitWidth = II->getType()->getIntegerBitWidth();
+      // If the length is 0 or the shift is out of range, replace with zero.
+      if (Length == 0 || Shift >= BitWidth)
+        return replaceInstUsesWith(CI, ConstantInt::get(II->getType(), 0));
+      // If the LHS is also a constant, we can completely constant fold this.
+      if (auto *InC = dyn_cast<ConstantInt>(II->getArgOperand(0))) {
+        uint64_t Result = InC->getZExtValue() >> Shift;
+        if (Length > BitWidth)
+          Length = BitWidth;
+        Result &= maskTrailingOnes<uint64_t>(Length);
+        return replaceInstUsesWith(CI, ConstantInt::get(II->getType(), Result));
+      }
+      // TODO should we turn this into 'and' if shift is 0? Or 'shl' if we
+      // are only masking bits that a shift already cleared?
+    }
+    break;
+
   case Intrinsic::x86_vcvtph2ps_128:
   case Intrinsic::x86_vcvtph2ps_256: {
     auto Arg = II->getArgOperand(0);
diff --git a/llvm/test/Transforms/InstCombine/X86/x86-bmi-tbm.ll b/llvm/test/Transforms/InstCombine/X86/x86-bmi-tbm.ll
new file mode 100644
index 00000000000..f090461fb04
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/X86/x86-bmi-tbm.ll
@@ -0,0 +1,203 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+declare i32 @llvm.x86.tbm.bextri.u32(i32, i32) nounwind readnone
+declare i64 @llvm.x86.tbm.bextri.u64(i64, i64) nounwind readnone
+declare i32 @llvm.x86.bmi.bextr.32(i32, i32) nounwind readnone
+declare i64 @llvm.x86.bmi.bextr.64(i64, i64) nounwind readnone
+
+define i32 @test_x86_tbm_bextri_u32(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 @llvm.x86.tbm.bextri.u32(i32 [[A:%.*]], i32 1296)
+; CHECK-NEXT:    ret i32 [[TMP1]]
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 %a, i32 1296)
+  ret i32 %1
+}
+
+define i32 @test_x86_tbm_bextri_u32_zero_length(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32_zero_length(
+; CHECK-NEXT:    ret i32 0
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 %a, i32 1)
+  ret i32 %1
+}
+
+define i32 @test_x86_tbm_bextri_u32_large_shift(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32_large_shift(
+; CHECK-NEXT:    ret i32 0
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 %a, i32 288)
+  ret i32 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.x86.tbm.bextri.u64(i64 [[A:%.*]], i64 1312)
+; CHECK-NEXT:    ret i64 [[TMP1]]
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 %a, i64 1312)
+  ret i64 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64_zero_length(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64_zero_length(
+; CHECK-NEXT:    ret i64 0
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 %a, i64 1)
+  ret i64 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64_large_shift(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64_large_shift(
+; CHECK-NEXT:    ret i64 0
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 %a, i64 320)
+  ret i64 %1
+}
+
+define i32 @test_x86_tbm_bextri_u32_constfold() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32_constfold(
+; CHECK-NEXT:    ret i32 57005
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 3735928559, i32 4112) ; extract bits 31:16 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i32 @test_x86_tbm_bextri_u32_constfold2() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32_constfold2(
+; CHECK-NEXT:    ret i32 233495534
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 3735928559, i32 8196) ; extract bits 35:4 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i32 @test_x86_tbm_bextri_u32_constfold3() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u32_constfold3(
+; CHECK-NEXT:    ret i32 233495534
+;
+  %1 = tail call i32 @llvm.x86.tbm.bextri.u32(i32 3735928559, i32 16388) ; extract bits 67:4 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64_constfold() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64_constfold(
+; CHECK-NEXT:    ret i64 57005
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 3735928559, i64 4112) ; extract bits 31:16 from 0xDEADBEEF
+  ret i64 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64_constfold2() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64_constfold2(
+; CHECK-NEXT:    ret i64 233495534
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 3735928559, i64 16388) ; extract bits 67:4 from 0xDEADBEEF
+  ret i64 %1
+}
+
+define i64 @test_x86_tbm_bextri_u64_constfold3() nounwind readnone {
+; CHECK-LABEL: @test_x86_tbm_bextri_u64_constfold3(
+; CHECK-NEXT:    ret i64 233495534
+;
+  %1 = tail call i64 @llvm.x86.tbm.bextri.u64(i64 3735928559, i64 32772) ; extract bits 131:4 from 0xDEADBEEF
+  ret i64 %1
+}
+
+define i32 @test_x86_bmi_bextri_32(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 @llvm.x86.bmi.bextr.32(i32 [[A:%.*]], i32 1296)
+; CHECK-NEXT:    ret i32 [[TMP1]]
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 %a, i32 1296)
+  ret i32 %1
+}
+
+define i32 @test_x86_bmi_bextri_32_zero_length(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32_zero_length(
+; CHECK-NEXT:    ret i32 0
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 %a, i32 1)
+  ret i32 %1
+}
+
+define i32 @test_x86_bmi_bextri_32_large_shift(i32 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32_large_shift(
+; CHECK-NEXT:    ret i32 0
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 %a, i32 288)
+  ret i32 %1
+}
+
+define i64 @test_x86_bmi_bextri_64(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i64 @llvm.x86.bmi.bextr.64(i64 [[A:%.*]], i64 1312)
+; CHECK-NEXT:    ret i64 [[TMP1]]
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 %a, i64 1312)
+  ret i64 %1
+}
+
+define i64 @test_x86_bmi_bextri_64_zero_length(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64_zero_length(
+; CHECK-NEXT:    ret i64 0
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 %a, i64 1)
+  ret i64 %1
+}
+
+define i64 @test_x86_bmi_bextri_64_large_shift(i64 %a) nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64_large_shift(
+; CHECK-NEXT:    ret i64 0
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 %a, i64 320)
+  ret i64 %1
+}
+
+define i32 @test_x86_bmi_bextri_32_constfold() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32_constfold(
+; CHECK-NEXT:    ret i32 57005
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 3735928559, i32 4112) ; extract bits 31:16 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i32 @test_x86_bmi_bextri_32_constfold2() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32_constfold2(
+; CHECK-NEXT:    ret i32 233495534
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 3735928559, i32 8196) ; extract bits 35:4 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i32 @test_x86_bmi_bextri_32_constfold3() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_32_constfold3(
+; CHECK-NEXT:    ret i32 233495534
+;
+  %1 = tail call i32 @llvm.x86.bmi.bextr.32(i32 3735928559, i32 16388) ; extract bits 67:4 from 0xDEADBEEF
+  ret i32 %1
+}
+
+define i64 @test_x86_bmi_bextri_64_constfold() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64_constfold(
+; CHECK-NEXT:    ret i64 57005
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 3735928559, i64 4112) ; extract bits 31:16 from 0xDEADBEEF
+  ret i64 %1
+}
+
+define i64 @test_x86_bmi_bextri_64_constfold2() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64_constfold2(
+; CHECK-NEXT:    ret i64 233495534
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 3735928559, i64 16388) ; extract bits 67:4 from 0xDEADBEEF
+  ret i64 %1
+}
+
+define i64 @test_x86_bmi_bextri_64_constfold3() nounwind readnone {
+; CHECK-LABEL: @test_x86_bmi_bextri_64_constfold3(
+; CHECK-NEXT:    ret i64 233495534
+;
+  %1 = tail call i64 @llvm.x86.bmi.bextr.64(i64 3735928559, i64 32772) ; extract bits 131:4 from 0xDEADBEEF
+  ret i64 %1
+}