From 9dada83d6c533376b919eaf93cea99772f180814 Mon Sep 17 00:00:00 2001 From: Sanjay Patel Date: Tue, 26 Feb 2019 18:26:56 +0000 Subject: [PATCH] [InstSimplify] remove zero-shift-guard fold for general funnel shift As discussed on llvm-dev: http://lists.llvm.org/pipermail/llvm-dev/2019-February/130491.html We can't remove the compare+select in the general case because we are treating funnel shift like a standard instruction (as opposed to a special instruction like select/phi). That means that if one of the operands of the funnel shift is poison, the result is poison regardless of whether we know that the operand is actually unused based on the instruction's particular semantics. The motivating case for this transform is the more specific rotate op (rather than funnel shift), and we are preserving the fold for that case because there is no chance of introducing extra poison when there is no anonymous extra operand to the funnel shift. llvm-svn: 354905 --- llvm/lib/Analysis/InstructionSimplify.cpp | 41 +++++++++++++++++++++--------- llvm/test/Transforms/InstSimplify/call.ll | 42 ++++++++++++++++++------------- 2 files changed, 54 insertions(+), 29 deletions(-) diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp index ed857dd..baf72a0 100644 --- a/llvm/lib/Analysis/InstructionSimplify.cpp +++ b/llvm/lib/Analysis/InstructionSimplify.cpp @@ -3908,27 +3908,44 @@ static Value *simplifySelectWithICmpCond(Value *CondVal, Value *TrueVal, Pred == ICmpInst::ICMP_EQ)) return V; - // Test for zero-shift-guard-ops around funnel shifts. These are used to - // avoid UB from oversized shifts in raw IR rotate patterns, but the - // intrinsics do not have that problem. + // Test for a bogus zero-shift-guard-op around funnel-shift or rotate. Value *ShAmt; auto isFsh = m_CombineOr(m_Intrinsic(m_Value(X), m_Value(), m_Value(ShAmt)), m_Intrinsic(m_Value(), m_Value(X), m_Value(ShAmt))); - // (ShAmt != 0) ? fshl(X, *, ShAmt) : X --> fshl(X, *, ShAmt) - // (ShAmt != 0) ? fshr(*, X, ShAmt) : X --> fshr(*, X, ShAmt) // (ShAmt == 0) ? fshl(X, *, ShAmt) : X --> X // (ShAmt == 0) ? fshr(*, X, ShAmt) : X --> X - if (match(TrueVal, isFsh) && FalseVal == X && CmpLHS == ShAmt) - return Pred == ICmpInst::ICMP_NE ? TrueVal : X; - - // (ShAmt == 0) ? X : fshl(X, *, ShAmt) --> fshl(X, *, ShAmt) - // (ShAmt == 0) ? X : fshr(*, X, ShAmt) --> fshr(*, X, ShAmt) + if (match(TrueVal, isFsh) && FalseVal == X && CmpLHS == ShAmt && + Pred == ICmpInst::ICMP_EQ) + return X; // (ShAmt != 0) ? X : fshl(X, *, ShAmt) --> X // (ShAmt != 0) ? X : fshr(*, X, ShAmt) --> X - if (match(FalseVal, isFsh) && TrueVal == X && CmpLHS == ShAmt) - return Pred == ICmpInst::ICMP_EQ ? FalseVal : X; + if (match(FalseVal, isFsh) && TrueVal == X && CmpLHS == ShAmt && + Pred == ICmpInst::ICMP_NE) + return X; + + // Test for a zero-shift-guard-op around rotates. These are used to + // avoid UB from oversized shifts in raw IR rotate patterns, but the + // intrinsics do not have that problem. + // We do not allow this transform for the general funnel shift case because + // that would not preserve the poison safety of the original code. + auto isRotate = m_CombineOr(m_Intrinsic(m_Value(X), + m_Deferred(X), + m_Value(ShAmt)), + m_Intrinsic(m_Value(X), + m_Deferred(X), + m_Value(ShAmt))); + // (ShAmt != 0) ? fshl(X, X, ShAmt) : X --> fshl(X, X, ShAmt) + // (ShAmt != 0) ? fshr(X, X, ShAmt) : X --> fshr(X, X, ShAmt) + if (match(TrueVal, isRotate) && FalseVal == X && CmpLHS == ShAmt && + Pred == ICmpInst::ICMP_NE) + return TrueVal; + // (ShAmt == 0) ? X : fshl(X, X, ShAmt) --> fshl(X, X, ShAmt) + // (ShAmt == 0) ? X : fshr(X, X, ShAmt) --> fshr(X, X, ShAmt) + if (match(FalseVal, isRotate) && TrueVal == X && CmpLHS == ShAmt && + Pred == ICmpInst::ICMP_EQ) + return FalseVal; } // Check for other compares that behave like bit test. diff --git a/llvm/test/Transforms/InstSimplify/call.ll b/llvm/test/Transforms/InstSimplify/call.ll index dfad3ef..0771e44 100644 --- a/llvm/test/Transforms/InstSimplify/call.ll +++ b/llvm/test/Transforms/InstSimplify/call.ll @@ -470,12 +470,14 @@ define <2 x i8> @fshr_no_shift_modulo_bitwidth_splat(<2 x i8> %x, <2 x i8> %y) { ret <2 x i8> %z } -; When the shift amount is 0, fshl returns its 1st parameter (x), so the guard is not needed. +; If y is poison, eliminating the guard is not safe. define i8 @fshl_zero_shift_guard(i8 %x, i8 %y, i8 %sh) { ; CHECK-LABEL: @fshl_zero_shift_guard( -; CHECK-NEXT: [[F:%.*]] = call i8 @llvm.fshl.i8(i8 [[X:%.*]], i8 [[Y:%.*]], i8 [[SH:%.*]]) -; CHECK-NEXT: ret i8 [[F]] +; CHECK-NEXT: [[C:%.*]] = icmp eq i8 [[SH:%.*]], 0 +; CHECK-NEXT: [[F:%.*]] = call i8 @llvm.fshl.i8(i8 [[X:%.*]], i8 [[Y:%.*]], i8 [[SH]]) +; CHECK-NEXT: [[S:%.*]] = select i1 [[C]], i8 [[X]], i8 [[F]] +; CHECK-NEXT: ret i8 [[S]] ; %c = icmp eq i8 %sh, 0 %f = call i8 @llvm.fshl.i8(i8 %x, i8 %y, i8 %sh) @@ -483,12 +485,14 @@ define i8 @fshl_zero_shift_guard(i8 %x, i8 %y, i8 %sh) { ret i8 %s } -; When the shift amount is 0, fshl returns its 1st parameter (x), so the guard is not needed. +; If y is poison, eliminating the guard is not safe. define i8 @fshl_zero_shift_guard_swapped(i8 %x, i8 %y, i8 %sh) { ; CHECK-LABEL: @fshl_zero_shift_guard_swapped( -; CHECK-NEXT: [[F:%.*]] = call i8 @llvm.fshl.i8(i8 [[X:%.*]], i8 [[Y:%.*]], i8 [[SH:%.*]]) -; CHECK-NEXT: ret i8 [[F]] +; CHECK-NEXT: [[C:%.*]] = icmp ne i8 [[SH:%.*]], 0 +; CHECK-NEXT: [[F:%.*]] = call i8 @llvm.fshl.i8(i8 [[X:%.*]], i8 [[Y:%.*]], i8 [[SH]]) +; CHECK-NEXT: [[S:%.*]] = select i1 [[C]], i8 [[F]], i8 [[X]] +; CHECK-NEXT: ret i8 [[S]] ; %c = icmp ne i8 %sh, 0 %f = call i8 @llvm.fshl.i8(i8 %x, i8 %y, i8 %sh) @@ -520,12 +524,14 @@ define i8 @fshl_zero_shift_guard_inverted_swapped(i8 %x, i8 %y, i8 %sh) { ret i8 %s } -; When the shift amount is 0, fshr returns its 2nd parameter (y), so the guard is not needed. +; If x is poison, eliminating the guard is not safe. define i9 @fshr_zero_shift_guard(i9 %x, i9 %y, i9 %sh) { ; CHECK-LABEL: @fshr_zero_shift_guard( -; CHECK-NEXT: [[F:%.*]] = call i9 @llvm.fshr.i9(i9 [[X:%.*]], i9 [[Y:%.*]], i9 [[SH:%.*]]) -; CHECK-NEXT: ret i9 [[F]] +; CHECK-NEXT: [[C:%.*]] = icmp eq i9 [[SH:%.*]], 0 +; CHECK-NEXT: [[F:%.*]] = call i9 @llvm.fshr.i9(i9 [[X:%.*]], i9 [[Y:%.*]], i9 [[SH]]) +; CHECK-NEXT: [[S:%.*]] = select i1 [[C]], i9 [[Y]], i9 [[F]] +; CHECK-NEXT: ret i9 [[S]] ; %c = icmp eq i9 %sh, 0 %f = call i9 @llvm.fshr.i9(i9 %x, i9 %y, i9 %sh) @@ -533,12 +539,14 @@ define i9 @fshr_zero_shift_guard(i9 %x, i9 %y, i9 %sh) { ret i9 %s } -; When the shift amount is 0, fshr returns its 2nd parameter (y), so the guard is not needed. +; If x is poison, eliminating the guard is not safe. define i9 @fshr_zero_shift_guard_swapped(i9 %x, i9 %y, i9 %sh) { ; CHECK-LABEL: @fshr_zero_shift_guard_swapped( -; CHECK-NEXT: [[F:%.*]] = call i9 @llvm.fshr.i9(i9 [[X:%.*]], i9 [[Y:%.*]], i9 [[SH:%.*]]) -; CHECK-NEXT: ret i9 [[F]] +; CHECK-NEXT: [[C:%.*]] = icmp ne i9 [[SH:%.*]], 0 +; CHECK-NEXT: [[F:%.*]] = call i9 @llvm.fshr.i9(i9 [[X:%.*]], i9 [[Y:%.*]], i9 [[SH]]) +; CHECK-NEXT: [[S:%.*]] = select i1 [[C]], i9 [[F]], i9 [[Y]] +; CHECK-NEXT: ret i9 [[S]] ; %c = icmp ne i9 %sh, 0 %f = call i9 @llvm.fshr.i9(i9 %x, i9 %y, i9 %sh) @@ -687,14 +695,14 @@ define i8 @fshl_zero_shift_guard_wrong_select_op(i8 %x, i8 %y, i8 %sh) { ; Vector types work too. -define <2 x i8> @fshr_zero_shift_guard_splat(<2 x i8> %x, <2 x i8> %y, <2 x i8> %sh) { -; CHECK-LABEL: @fshr_zero_shift_guard_splat( -; CHECK-NEXT: [[F:%.*]] = call <2 x i8> @llvm.fshr.v2i8(<2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <2 x i8> [[SH:%.*]]) +define <2 x i8> @rotr_zero_shift_guard_splat(<2 x i8> %x, <2 x i8> %sh) { +; CHECK-LABEL: @rotr_zero_shift_guard_splat( +; CHECK-NEXT: [[F:%.*]] = call <2 x i8> @llvm.fshr.v2i8(<2 x i8> [[X:%.*]], <2 x i8> [[X]], <2 x i8> [[SH:%.*]]) ; CHECK-NEXT: ret <2 x i8> [[F]] ; %c = icmp eq <2 x i8> %sh, zeroinitializer - %f = call <2 x i8> @llvm.fshr.v2i8(<2 x i8> %x, <2 x i8> %y, <2 x i8> %sh) - %s = select <2 x i1> %c, <2 x i8> %y, <2 x i8> %f + %f = call <2 x i8> @llvm.fshr.v2i8(<2 x i8> %x, <2 x i8> %x, <2 x i8> %sh) + %s = select <2 x i1> %c, <2 x i8> %x, <2 x i8> %f ret <2 x i8> %s } -- 2.7.4