From: Roman Lebedev Date: Mon, 16 Jul 2018 16:45:42 +0000 (+0000) Subject: [InstCombine] Fold 'check for [no] signed truncation' pattern X-Git-Tag: android-x86-8.1-r1~753 X-Git-Url: http://git.osdn.net/view?a=commitdiff_plain;h=a5425a350ebd4b13d4b74616133c6fc0a74c4598;p=android-x86%2Fexternal-llvm.git [InstCombine] Fold 'check for [no] signed truncation' pattern Summary: [[ https://bugs.llvm.org/show_bug.cgi?id=38149 | PR38149 ]] As discussed in https://reviews.llvm.org/D49179#1158957 and later, the IR for 'check for [no] signed truncation' pattern can be improved: https://rise4fun.com/Alive/gBf ^ that pattern will be produced by Implicit Integer Truncation sanitizer, https://reviews.llvm.org/D48958 https://bugs.llvm.org/show_bug.cgi?id=21530 in signed case, therefore it is probably a good idea to improve it. Proofs for this transform: https://rise4fun.com/Alive/mgu This transform is surprisingly frustrating. This does not deal with non-splat shift amounts, or with undef shift amounts. I've outlined what i think the solution should be: ``` // Potential handling of non-splats: for each element: // * if both are undef, replace with constant 0. // Because (1<<0) is OK and is 1, and ((1<<0)>>1) is also OK and is 0. // * if both are not undef, and are different, bailout. // * else, only one is undef, then pick the non-undef one. ``` The DAGCombine will reverse this transform, see https://reviews.llvm.org/D49266 Reviewers: spatel, craig.topper Reviewed By: spatel Subscribers: JDevlieghere, rkruppe, llvm-commits Differential Revision: https://reviews.llvm.org/D49320 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337190 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Transforms/InstCombine/InstCombineCompares.cpp b/lib/Transforms/InstCombine/InstCombineCompares.cpp index f91eb9da99e..d50dfa75552 100644 --- a/lib/Transforms/InstCombine/InstCombineCompares.cpp +++ b/lib/Transforms/InstCombine/InstCombineCompares.cpp @@ -2945,6 +2945,72 @@ static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I, return Builder.CreateICmp(DstPred, X, M); } +/// Some comparisons can be simplified. +/// In this case, we are looking for comparisons that look like +/// a check for a lossy signed truncation. +/// Folds: (MaskedBits is a constant.) +/// ((%x << MaskedBits) a>> MaskedBits) SrcPred %x +/// Into: +/// (add %x, (1 << (KeptBits-1))) DstPred (1 << KeptBits) +/// Where KeptBits = bitwidth(%x) - MaskedBits +static Value * +foldICmpWithTruncSignExtendedVal(ICmpInst &I, + InstCombiner::BuilderTy &Builder) { + ICmpInst::Predicate SrcPred; + Value *X; + const APInt *C0, *C1; // FIXME: non-splats, potentially with undef. + // We are ok with 'shl' having multiple uses, but 'ashr' must be one-use. + if (!match(&I, m_c_ICmp(SrcPred, + m_OneUse(m_AShr(m_Shl(m_Value(X), m_APInt(C0)), + m_APInt(C1))), + m_Deferred(X)))) + return nullptr; + + // Potential handling of non-splats: for each element: + // * if both are undef, replace with constant 0. + // Because (1<<0) is OK and is 1, and ((1<<0)>>1) is also OK and is 0. + // * if both are not undef, and are different, bailout. + // * else, only one is undef, then pick the non-undef one. + + // The shift amount must be equal. + if (*C0 != *C1) + return nullptr; + const uint64_t MaskedBits = C0->getZExtValue(); + assert(MaskedBits && "shift of %x by zero should be folded to %x already."); + + ICmpInst::Predicate DstPred; + switch (SrcPred) { + case ICmpInst::Predicate::ICMP_EQ: + // ((%x << MaskedBits) a>> MaskedBits) == %x + // => + // (add %x, (1 << (KeptBits-1))) u< (1 << KeptBits) + DstPred = ICmpInst::Predicate::ICMP_ULT; + break; + case ICmpInst::Predicate::ICMP_NE: + // ((%x << MaskedBits) a>> MaskedBits) != %x + // => + // (add %x, (1 << (KeptBits-1))) u>= (1 << KeptBits) + DstPred = ICmpInst::Predicate::ICMP_UGE; + break; + // FIXME: are more folds possible? + default: + return nullptr; + } + + const uint64_t XBitWidth = C0->getBitWidth(); + const uint64_t KeptBits = XBitWidth - MaskedBits; + const uint64_t ICmpCst = 1UL << KeptBits; // (1 << KeptBits) + const uint64_t AddCst = ICmpCst >> 1UL; // (1 << (KeptBits-1)) + + auto *XType = X->getType(); + // (add %x, (1 << (KeptBits-1))) + Value *T0 = Builder.CreateAdd(X, ConstantInt::get(XType, AddCst)); + // add %x, (1 << (KeptBits-1))) DstPred (1 << KeptBits) + Value *T1 = Builder.CreateICmp(DstPred, T0, ConstantInt::get(XType, ICmpCst)); + + return T1; +} + /// Try to fold icmp (binop), X or icmp X, (binop). /// TODO: A large part of this logic is duplicated in InstSimplify's /// simplifyICmpWithBinOp(). We should be able to share that and avoid the code @@ -3285,6 +3351,9 @@ Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) { if (Value *V = foldICmpWithLowBitMaskedVal(I, Builder)) return replaceInstUsesWith(I, V); + if (Value *V = foldICmpWithTruncSignExtendedVal(I, Builder)) + return replaceInstUsesWith(I, V); + return nullptr; } diff --git a/test/Transforms/InstCombine/canonicalize-lack-of-signed-truncation-check.ll b/test/Transforms/InstCombine/canonicalize-lack-of-signed-truncation-check.ll index 4f324f9cd1e..306d3c917c8 100644 --- a/test/Transforms/InstCombine/canonicalize-lack-of-signed-truncation-check.ll +++ b/test/Transforms/InstCombine/canonicalize-lack-of-signed-truncation-check.ll @@ -15,9 +15,8 @@ define i1 @p0(i8 %x) { ; CHECK-LABEL: @p0( -; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X:%.*]], 4 -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i8 [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X:%.*]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ult i8 [[TMP1]], 16 ; CHECK-NEXT: ret i1 [[TMP2]] ; %tmp0 = shl i8 %x, 4 @@ -32,9 +31,8 @@ define i1 @p0(i8 %x) { define <2 x i1> @p1_vec_splat(<2 x i8> %x) { ; CHECK-LABEL: @p1_vec_splat( -; CHECK-NEXT: [[TMP0:%.*]] = shl <2 x i8> [[X:%.*]], -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact <2 x i8> [[TMP0]], -; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <2 x i8> [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], +; CHECK-NEXT: [[TMP2:%.*]] = icmp ult <2 x i8> [[TMP1]], ; CHECK-NEXT: ret <2 x i1> [[TMP2]] ; %tmp0 = shl <2 x i8> %x, @@ -104,9 +102,8 @@ declare i8 @gen8() define i1 @c0() { ; CHECK-LABEL: @c0( ; CHECK-NEXT: [[X:%.*]] = call i8 @gen8() -; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X]], 4 -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i8 [[X]], [[TMP1]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ult i8 [[TMP1]], 16 ; CHECK-NEXT: ret i1 [[TMP2]] ; %x = call i8 @gen8() @@ -126,8 +123,8 @@ define i1 @n_oneuse0(i8 %x) { ; CHECK-LABEL: @n_oneuse0( ; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X:%.*]], 4 ; CHECK-NEXT: call void @use8(i8 [[TMP0]]) -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i8 [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ult i8 [[TMP1]], 16 ; CHECK-NEXT: ret i1 [[TMP2]] ; %tmp0 = shl i8 %x, 4 diff --git a/test/Transforms/InstCombine/canonicalize-signed-truncation-check.ll b/test/Transforms/InstCombine/canonicalize-signed-truncation-check.ll index f1a5034e26f..5a3345cf898 100644 --- a/test/Transforms/InstCombine/canonicalize-signed-truncation-check.ll +++ b/test/Transforms/InstCombine/canonicalize-signed-truncation-check.ll @@ -15,9 +15,8 @@ define i1 @p0(i8 %x) { ; CHECK-LABEL: @p0( -; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X:%.*]], 4 -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp ne i8 [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X:%.*]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i8 [[TMP1]], 15 ; CHECK-NEXT: ret i1 [[TMP2]] ; %tmp0 = shl i8 %x, 4 @@ -32,9 +31,8 @@ define i1 @p0(i8 %x) { define <2 x i1> @p1_vec_splat(<2 x i8> %x) { ; CHECK-LABEL: @p1_vec_splat( -; CHECK-NEXT: [[TMP0:%.*]] = shl <2 x i8> [[X:%.*]], -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact <2 x i8> [[TMP0]], -; CHECK-NEXT: [[TMP2:%.*]] = icmp ne <2 x i8> [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add <2 x i8> [[X:%.*]], +; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <2 x i8> [[TMP1]], ; CHECK-NEXT: ret <2 x i1> [[TMP2]] ; %tmp0 = shl <2 x i8> %x, @@ -104,9 +102,8 @@ declare i8 @gen8() define i1 @c0() { ; CHECK-LABEL: @c0( ; CHECK-NEXT: [[X:%.*]] = call i8 @gen8() -; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X]], 4 -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp ne i8 [[X]], [[TMP1]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i8 [[TMP1]], 15 ; CHECK-NEXT: ret i1 [[TMP2]] ; %x = call i8 @gen8() @@ -126,8 +123,8 @@ define i1 @n_oneuse0(i8 %x) { ; CHECK-LABEL: @n_oneuse0( ; CHECK-NEXT: [[TMP0:%.*]] = shl i8 [[X:%.*]], 4 ; CHECK-NEXT: call void @use8(i8 [[TMP0]]) -; CHECK-NEXT: [[TMP1:%.*]] = ashr exact i8 [[TMP0]], 4 -; CHECK-NEXT: [[TMP2:%.*]] = icmp ne i8 [[TMP1]], [[X]] +; CHECK-NEXT: [[TMP1:%.*]] = add i8 [[X]], 8 +; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i8 [[TMP1]], 15 ; CHECK-NEXT: ret i1 [[TMP2]] ; %tmp0 = shl i8 %x, 4