diff options
| author | AtariDreams <gfunni234@gmail.com> | 2024-05-30 03:39:32 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-05-30 09:39:32 +0200 |
| commit | 1034b4d38dd2df542204d55b3e14e985452e4fe4 (patch) | |
| tree | 85091cfa250e4a72187adfffd91765268e4e65f1 | |
| parent | 0eb4bf2faf4125d4d279463390a753c8c36a6937 (diff) | |
| download | llvm-1034b4d38dd2df542204d55b3e14e985452e4fe4.zip llvm-1034b4d38dd2df542204d55b3e14e985452e4fe4.tar.gz llvm-1034b4d38dd2df542204d55b3e14e985452e4fe4.tar.bz2 | |
[InstCombine] lshr (mul (X, 2^N + 1)), N -> X when X is half-width (#93677)
Alive2 Proof:
https://alive2.llvm.org/ce/z/Yd2CKF
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp | 6 | ||||
| -rw-r--r-- | llvm/test/Transforms/InstCombine/lshr.ll | 10 |
2 files changed, 6 insertions, 10 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp index 4f91993..fbc02cddf 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp @@ -1464,10 +1464,10 @@ Instruction *InstCombinerImpl::visitLShr(BinaryOperator &I) { if (BitWidth > 2 && (*MulC - 1).isPowerOf2() && MulC->logBase2() == ShAmtC) { // Look for a "splat" mul pattern - it replicates bits across each half - // of a value, so a right shift is just a mask of the low bits: - // lshr i[2N] (mul nuw X, (2^N)+1), N --> and iN X, (2^N)-1 + // of a value, so a right shift simplifies back to just X: + // lshr i[2N] (mul nuw X, (2^N)+1), N --> X if (ShAmtC * 2 == BitWidth) - return BinaryOperator::CreateAnd(X, ConstantInt::get(Ty, *MulC - 2)); + return replaceInstUsesWith(I, X); // lshr (mul nuw (X, 2^N + 1)), N -> add nuw (X, lshr(X, N)) if (Op0->hasOneUse()) { diff --git a/llvm/test/Transforms/InstCombine/lshr.ll b/llvm/test/Transforms/InstCombine/lshr.ll index dfdb6c7..0392764 100644 --- a/llvm/test/Transforms/InstCombine/lshr.ll +++ b/llvm/test/Transforms/InstCombine/lshr.ll @@ -348,8 +348,7 @@ define <2 x i32> @narrow_lshr_constant(<2 x i8> %x, <2 x i8> %y) { define i32 @mul_splat_fold(i32 %x) { ; CHECK-LABEL: @mul_splat_fold( -; CHECK-NEXT: [[T:%.*]] = and i32 [[X:%.*]], 65535 -; CHECK-NEXT: ret i32 [[T]] +; CHECK-NEXT: ret i32 [[X:%.*]] ; %m = mul nuw i32 %x, 65537 %t = lshr i32 %m, 16 @@ -362,8 +361,7 @@ define <3 x i14> @mul_splat_fold_vec(<3 x i14> %x) { ; CHECK-LABEL: @mul_splat_fold_vec( ; CHECK-NEXT: [[M:%.*]] = mul nuw <3 x i14> [[X:%.*]], <i14 129, i14 129, i14 129> ; CHECK-NEXT: call void @usevec(<3 x i14> [[M]]) -; CHECK-NEXT: [[T:%.*]] = and <3 x i14> [[X]], <i14 127, i14 127, i14 127> -; CHECK-NEXT: ret <3 x i14> [[T]] +; CHECK-NEXT: ret <3 x i14> [[X]] ; %m = mul nuw <3 x i14> %x, <i14 129, i14 129, i14 129> call void @usevec(<3 x i14> %m) @@ -628,8 +626,6 @@ define i32 @mul_splat_fold_wrong_lshr_const(i32 %x) { ret i32 %t } -; Negative test (but simplifies into a different transform) - define i32 @mul_splat_fold_no_nuw(i32 %x) { ; CHECK-LABEL: @mul_splat_fold_no_nuw( ; CHECK-NEXT: [[TMP1:%.*]] = lshr i32 [[X:%.*]], 16 @@ -641,7 +637,7 @@ define i32 @mul_splat_fold_no_nuw(i32 %x) { ret i32 %t } -; Negative test +; Negative test define i32 @mul_splat_fold_no_flags(i32 %x) { ; CHECK-LABEL: @mul_splat_fold_no_flags( |