[llvm] r336585 - [X86][TLI] DAGCombine: Unfold variable bit-clearing mask to two shifts.
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 9 12:06:42 PDT 2018
Author: lebedevri
Date: Mon Jul 9 12:06:42 2018
New Revision: 336585
URL: http://llvm.org/viewvc/llvm-project?rev=336585&view=rev
Log:
[X86][TLI] DAGCombine: Unfold variable bit-clearing mask to two shifts.
Summary:
This adds a reverse transform for the instcombine canonicalizations
that were added in D47980, D47981.
As discussed later, that was worse at least for the code size,
and potentially for the performance, too.
https://rise4fun.com/Alive/Zmpl
Reviewers: craig.topper, RKSimon, spatel
Reviewed By: spatel
Subscribers: reames, llvm-commits
Differential Revision: https://reviews.llvm.org/D48768
Modified:
llvm/trunk/include/llvm/CodeGen/TargetLowering.h
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/lib/Target/X86/X86ISelLowering.h
llvm/trunk/test/CodeGen/X86/clear-highbits.ll
llvm/trunk/test/CodeGen/X86/clear-lowbits.ll
llvm/trunk/test/CodeGen/X86/extract-lowbits.ll
Modified: llvm/trunk/include/llvm/CodeGen/TargetLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/TargetLowering.h?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/include/llvm/CodeGen/TargetLowering.h (original)
+++ llvm/trunk/include/llvm/CodeGen/TargetLowering.h Mon Jul 9 12:06:42 2018
@@ -509,6 +509,16 @@ public:
return hasAndNotCompare(X);
}
+ /// There are two ways to clear extreme bits (either low or high):
+ /// Mask: x & (-1 << y) (the instcombine canonical form)
+ /// Shifts: x >> y << y
+ /// Return true if the variant with 2 shifts is preferred.
+ /// Return false if there is no preference.
+ virtual bool preferShiftsToClearExtremeBits(SDValue X) const {
+ // By default, let's assume that no one prefers shifts.
+ return false;
+ }
+
/// Return true if the target wants to use the optimization that
/// turns ext(promotableInst1(...(promotableInstN(load)))) into
/// promotedInst1(...(promotedInstN(ext(load)))).
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Mon Jul 9 12:06:42 2018
@@ -409,6 +409,7 @@ namespace {
SDValue foldLogicOfSetCCs(bool IsAnd, SDValue N0, SDValue N1,
const SDLoc &DL);
SDValue unfoldMaskedMerge(SDNode *N);
+ SDValue unfoldExtremeBitClearingToShifts(SDNode *N);
SDValue SimplifySetCC(EVT VT, SDValue N0, SDValue N1, ISD::CondCode Cond,
const SDLoc &DL, bool foldBooleans);
SDValue rebuildSetCC(SDValue N);
@@ -4169,6 +4170,60 @@ bool DAGCombiner::BackwardsPropagateMask
return false;
}
+// Unfold
+// x & (-1 'logical shift' y)
+// To
+// (x 'opposite logical shift' y) 'logical shift' y
+// if it is better for performance.
+SDValue DAGCombiner::unfoldExtremeBitClearingToShifts(SDNode *N) {
+ assert(N->getOpcode() == ISD::AND);
+
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+
+ // Do we actually prefer shifts over mask?
+ if (!TLI.preferShiftsToClearExtremeBits(N0))
+ return SDValue();
+
+ // Try to match (-1 '[outer] logical shift' y)
+ unsigned OuterShift;
+ unsigned InnerShift; // The opposite direction to the OuterShift.
+ SDValue Y; // Shift amount.
+ auto matchMask = [&OuterShift, &InnerShift, &Y](SDValue M) -> bool {
+ if (!M.hasOneUse())
+ return false;
+ OuterShift = M->getOpcode();
+ if (OuterShift == ISD::SHL)
+ InnerShift = ISD::SRL;
+ else if (OuterShift == ISD::SRL)
+ InnerShift = ISD::SHL;
+ else
+ return false;
+ if (!isAllOnesConstant(M->getOperand(0)))
+ return false;
+ Y = M->getOperand(1);
+ return true;
+ };
+
+ SDValue X;
+ if (matchMask(N1))
+ X = N0;
+ else if (matchMask(N0))
+ X = N1;
+ else
+ return SDValue();
+
+ SDLoc DL(N);
+ EVT VT = N->getValueType(0);
+
+ // tmp = x 'opposite logical shift' y
+ SDValue T0 = DAG.getNode(InnerShift, DL, VT, X, Y);
+ // ret = tmp 'logical shift' y
+ SDValue T1 = DAG.getNode(OuterShift, DL, VT, T0, Y);
+
+ return T1;
+}
+
SDValue DAGCombiner::visitAND(SDNode *N) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
@@ -4466,6 +4521,9 @@ SDValue DAGCombiner::visitAND(SDNode *N)
return BSwap;
}
+ if (SDValue Shifts = unfoldExtremeBitClearingToShifts(N))
+ return Shifts;
+
return SDValue();
}
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Mon Jul 9 12:06:42 2018
@@ -4786,6 +4786,20 @@ bool X86TargetLowering::hasAndNot(SDValu
return Subtarget.hasSSE2();
}
+bool X86TargetLowering::preferShiftsToClearExtremeBits(SDValue Y) const {
+ EVT VT = Y.getValueType();
+
+ // For vectors, we don't have a preference, but we probably want a mask.
+ if (VT.isVector())
+ return false;
+
+ // 64-bit shifts on 32-bit targets produce really bad bloated code.
+ if (VT == MVT::i64 && !Subtarget.is64Bit())
+ return false;
+
+ return true;
+}
+
MVT X86TargetLowering::hasFastEqualityCompare(unsigned NumBits) const {
MVT VT = MVT::getIntegerVT(NumBits);
if (isTypeLegal(VT))
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.h?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.h (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.h Mon Jul 9 12:06:42 2018
@@ -828,6 +828,8 @@ namespace llvm {
bool hasAndNot(SDValue Y) const override;
+ bool preferShiftsToClearExtremeBits(SDValue Y) const override;
+
bool convertSetCCLogicToBitwiseLogic(EVT VT) const override {
return VT.isScalarInteger();
}
Modified: llvm/trunk/test/CodeGen/X86/clear-highbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/clear-highbits.ll?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/clear-highbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/clear-highbits.ll Mon Jul 9 12:06:42 2018
@@ -25,17 +25,18 @@ define i8 @clear_highbits8_c0(i8 %val, i
; X86-LABEL: clear_highbits8_c0:
; X86: # %bb.0:
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: shlb %cl, %al
; X86-NEXT: shrb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_highbits8_c0:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
-; X64-NEXT: shrb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%mask = lshr i8 -1, %numhighbits
%masked = and i8 %mask, %val
@@ -45,19 +46,20 @@ define i8 @clear_highbits8_c0(i8 %val, i
define i8 @clear_highbits8_c2_load(i8* %w, i8 %numhighbits) nounwind {
; X86-LABEL: clear_highbits8_c2_load:
; X86: # %bb.0:
-; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movb (%eax), %al
+; X86-NEXT: shlb %cl, %al
; X86-NEXT: shrb %cl, %al
-; X86-NEXT: andb (%edx), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_highbits8_c2_load:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
+; X64-NEXT: movb (%rdi), %al
+; X64-NEXT: shlb %cl, %al
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NEXT: shrb %cl, %al
-; X64-NEXT: andb (%rdi), %al
; X64-NEXT: retq
%val = load i8, i8* %w
%mask = lshr i8 -1, %numhighbits
@@ -69,17 +71,18 @@ define i8 @clear_highbits8_c4_commutativ
; X86-LABEL: clear_highbits8_c4_commutative:
; X86: # %bb.0:
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: shlb %cl, %al
; X86-NEXT: shrb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_highbits8_c4_commutative:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
-; X64-NEXT: shrb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%mask = lshr i8 -1, %numhighbits
%masked = and i8 %val, %mask ; swapped order
@@ -94,35 +97,37 @@ define i16 @clear_highbits16_c0(i16 %val
; X86-NOBMI2-LABEL: clear_highbits16_c0:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
+; X86-NOBMI2-NEXT: movzwl %ax, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits16_c0:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl %cx, %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits16_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits16_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT: movzwl %ax, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%mask = lshr i16 -1, %numhighbits
@@ -134,35 +139,37 @@ define i16 @clear_highbits16_c1_indexzex
; X86-NOBMI2-LABEL: clear_highbits16_c1_indexzext:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
+; X86-NOBMI2-NEXT: movzwl %ax, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits16_c1_indexzext:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl %cx, %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits16_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits16_c1_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT: movzwl %ax, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numhighbits to i16
@@ -174,38 +181,43 @@ define i16 @clear_highbits16_c1_indexzex
define i16 @clear_highbits16_c2_load(i16* %w, i16 %numhighbits) nounwind {
; X86-NOBMI2-LABEL: clear_highbits16_c2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
+; X86-NOBMI2-NEXT: movzwl %ax, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits16_c2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $65535, %edx # imm = 0xFFFF
-; X86-BMI2-NEXT: shrxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl (%ecx), %ecx
+; X86-BMI2-NEXT: shlxl %eax, %ecx, %ecx
+; X86-BMI2-NEXT: movzwl %cx, %ecx
+; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits16_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
+; X64-NOBMI2-NEXT: shll %cl, %eax
+; X64-NOBMI2-NEXT: movzwl %ax, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits16_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT: movzwl %ax, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -217,38 +229,43 @@ define i16 @clear_highbits16_c2_load(i16
define i16 @clear_highbits16_c3_load_indexzext(i16* %w, i8 %numhighbits) nounwind {
; X86-NOBMI2-LABEL: clear_highbits16_c3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
+; X86-NOBMI2-NEXT: movzwl %ax, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits16_c3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $65535, %edx # imm = 0xFFFF
-; X86-BMI2-NEXT: shrxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl (%ecx), %ecx
+; X86-BMI2-NEXT: shlxl %eax, %ecx, %ecx
+; X86-BMI2-NEXT: movzwl %cx, %ecx
+; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits16_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
+; X64-NOBMI2-NEXT: shll %cl, %eax
+; X64-NOBMI2-NEXT: movzwl %ax, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits16_c3_load_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
+; X64-BMI2-NEXT: movzwl %ax, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -262,35 +279,37 @@ define i16 @clear_highbits16_c4_commutat
; X86-NOBMI2-LABEL: clear_highbits16_c4_commutative:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
+; X86-NOBMI2-NEXT: movzwl %ax, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits16_c4_commutative:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $65535, %ecx # imm = 0xFFFF
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl %cx, %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits16_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits16_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $65535, %eax # imm = 0xFFFF
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
+; X64-BMI2-NEXT: movzwl %ax, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%mask = lshr i16 -1, %numhighbits
@@ -306,32 +325,31 @@ define i32 @clear_highbits32_c0(i32 %val
; X86-NOBMI2-LABEL: clear_highbits32_c0:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits32_c0:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits32_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits32_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%mask = lshr i32 -1, %numhighbits
%masked = and i32 %mask, %val
@@ -342,32 +360,31 @@ define i32 @clear_highbits32_c1_indexzex
; X86-NOBMI2-LABEL: clear_highbits32_c1_indexzext:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits32_c1_indexzext:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits32_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits32_c1_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numhighbits to i32
%mask = lshr i32 -1, %sh_prom
@@ -378,35 +395,34 @@ define i32 @clear_highbits32_c1_indexzex
define i32 @clear_highbits32_c2_load(i32* %w, i32 %numhighbits) nounwind {
; X86-NOBMI2-LABEL: clear_highbits32_c2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits32_c2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shrxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shlxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits32_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
+; X64-NOBMI2-NEXT: shll %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits32_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shlxl %esi, (%rdi), %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%mask = lshr i32 -1, %numhighbits
@@ -417,35 +433,34 @@ define i32 @clear_highbits32_c2_load(i32
define i32 @clear_highbits32_c3_load_indexzext(i32* %w, i8 %numhighbits) nounwind {
; X86-NOBMI2-LABEL: clear_highbits32_c3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits32_c3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shrxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shlxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits32_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
+; X64-NOBMI2-NEXT: shll %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits32_c3_load_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shlxl %esi, (%rdi), %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%sh_prom = zext i8 %numhighbits to i32
@@ -458,32 +473,31 @@ define i32 @clear_highbits32_c4_commutat
; X86-NOBMI2-LABEL: clear_highbits32_c4_commutative:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shll %cl, %eax
; X86-NOBMI2-NEXT: shrl %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_highbits32_c4_commutative:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shrxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_highbits32_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrl %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits32_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shlxl %esi, %edi, %eax
; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%mask = lshr i32 -1, %numhighbits
%masked = and i32 %val, %mask ; swapped order
@@ -530,17 +544,17 @@ define i64 @clear_highbits64_c0(i64 %val
;
; X64-NOBMI2-LABEL: clear_highbits64_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits64_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shlxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shrxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%mask = lshr i64 -1, %numhighbits
%masked = and i64 %mask, %val
@@ -583,18 +597,18 @@ define i64 @clear_highbits64_c1_indexzex
;
; X64-NOBMI2-LABEL: clear_highbits64_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits64_c1_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: # kill: def $esi killed $esi def $rsi
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shlxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shrxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numhighbits to i64
%mask = lshr i64 -1, %sh_prom
@@ -644,17 +658,17 @@ define i64 @clear_highbits64_c2_load(i64
;
; X64-NOBMI2-LABEL: clear_highbits64_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
+; X64-NOBMI2-NEXT: shlq %cl, %rax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT: shrq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits64_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shlxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT: shrxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%mask = lshr i64 -1, %numhighbits
@@ -704,18 +718,18 @@ define i64 @clear_highbits64_c3_load_ind
;
; X64-NOBMI2-LABEL: clear_highbits64_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
+; X64-NOBMI2-NEXT: shlq %cl, %rax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shrq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits64_c3_load_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: # kill: def $esi killed $esi def $rsi
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shlxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT: shrxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%sh_prom = zext i8 %numhighbits to i64
@@ -760,17 +774,17 @@ define i64 @clear_highbits64_c4_commutat
;
; X64-NOBMI2-LABEL: clear_highbits64_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shrq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_highbits64_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shlxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shrxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%mask = lshr i64 -1, %numhighbits
%masked = and i64 %val, %mask ; swapped order
Modified: llvm/trunk/test/CodeGen/X86/clear-lowbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/clear-lowbits.ll?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/clear-lowbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/clear-lowbits.ll Mon Jul 9 12:06:42 2018
@@ -27,17 +27,18 @@ define i8 @clear_lowbits8_c0(i8 %val, i8
; X86-LABEL: clear_lowbits8_c0:
; X86: # %bb.0:
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_c0:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
-; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%mask = shl i8 -1, %numlowbits
%masked = and i8 %mask, %val
@@ -47,19 +48,20 @@ define i8 @clear_lowbits8_c0(i8 %val, i8
define i8 @clear_lowbits8_c2_load(i8* %w, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_c2_load:
; X86: # %bb.0:
-; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movb (%eax), %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb (%edx), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_c2_load:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
+; X64-NEXT: movb (%rdi), %al
+; X64-NEXT: shrb %cl, %al
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb (%rdi), %al
; X64-NEXT: retq
%val = load i8, i8* %w
%mask = shl i8 -1, %numlowbits
@@ -71,17 +73,18 @@ define i8 @clear_lowbits8_c4_commutative
; X86-LABEL: clear_lowbits8_c4_commutative:
; X86: # %bb.0:
; X86-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_c4_commutative:
; X64: # %bb.0:
-; X64-NEXT: movb $-1, %al
; X64-NEXT: movl %esi, %ecx
-; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%mask = shl i8 -1, %numlowbits
%masked = and i8 %val, %mask ; swapped order
@@ -94,35 +97,36 @@ define i16 @clear_lowbits16_c0(i16 %val,
; X86-NOBMI2-LABEL: clear_lowbits16_c0:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_c0:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%mask = shl i16 -1, %numlowbits
@@ -134,35 +138,36 @@ define i16 @clear_lowbits16_c1_indexzext
; X86-NOBMI2-LABEL: clear_lowbits16_c1_indexzext:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_c1_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_c1_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numlowbits to i16
@@ -174,38 +179,39 @@ define i16 @clear_lowbits16_c1_indexzext
define i16 @clear_lowbits16_c2_load(i16* %w, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_c2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl (%ecx), %ecx
+; X86-BMI2-NEXT: shrxl %eax, %ecx, %ecx
+; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -217,38 +223,39 @@ define i16 @clear_lowbits16_c2_load(i16*
define i16 @clear_lowbits16_c3_load_indexzext(i16* %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: movzwl (%ecx), %ecx
+; X86-BMI2-NEXT: shrxl %eax, %ecx, %ecx
+; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_c3_load_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -262,35 +269,36 @@ define i16 @clear_lowbits16_c4_commutati
; X86-NOBMI2-LABEL: clear_lowbits16_c4_commutative:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_c4_commutative:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movzwl %di, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: shrxl %esi, %eax, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%mask = shl i16 -1, %numlowbits
@@ -304,32 +312,31 @@ define i32 @clear_lowbits32_c0(i32 %val,
; X86-NOBMI2-LABEL: clear_lowbits32_c0:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_c0:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shrxl %esi, %edi, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%mask = shl i32 -1, %numlowbits
%masked = and i32 %mask, %val
@@ -340,32 +347,31 @@ define i32 @clear_lowbits32_c1_indexzext
; X86-NOBMI2-LABEL: clear_lowbits32_c1_indexzext:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_c1_indexzext:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_c1_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shrxl %esi, %edi, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numlowbits to i32
%mask = shl i32 -1, %sh_prom
@@ -376,35 +382,34 @@ define i32 @clear_lowbits32_c1_indexzext
define i32 @clear_lowbits32_c2_load(i32* %w, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_c2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%mask = shl i32 -1, %numlowbits
@@ -415,35 +420,34 @@ define i32 @clear_lowbits32_c2_load(i32*
define i32 @clear_lowbits32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_c3_load_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shrxl %esi, (%rdi), %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%sh_prom = zext i8 %numlowbits to i32
@@ -456,32 +460,31 @@ define i32 @clear_lowbits32_c4_commutati
; X86-NOBMI2-LABEL: clear_lowbits32_c4_commutative:
; X86-NOBMI2: # %bb.0:
; X86-NOBMI2-NEXT: movb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_c4_commutative:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movl $-1, %eax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $-1, %eax
+; X64-BMI2-NEXT: shrxl %esi, %edi, %eax
; X64-BMI2-NEXT: shlxl %esi, %eax, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%mask = shl i32 -1, %numlowbits
%masked = and i32 %val, %mask ; swapped order
@@ -526,17 +529,17 @@ define i64 @clear_lowbits64_c0(i64 %val,
;
; X64-NOBMI2-LABEL: clear_lowbits64_c0:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_c0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%mask = shl i64 -1, %numlowbits
%masked = and i64 %mask, %val
@@ -579,18 +582,18 @@ define i64 @clear_lowbits64_c1_indexzext
;
; X64-NOBMI2-LABEL: clear_lowbits64_c1_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_c1_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: # kill: def $esi killed $esi def $rsi
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%sh_prom = zext i8 %numlowbits to i64
%mask = shl i64 -1, %sh_prom
@@ -640,17 +643,17 @@ define i64 @clear_lowbits64_c2_load(i64*
;
; X64-NOBMI2-LABEL: clear_lowbits64_c2_load:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_c2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%mask = shl i64 -1, %numlowbits
@@ -700,18 +703,18 @@ define i64 @clear_lowbits64_c3_load_inde
;
; X64-NOBMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
; X64-NOBMI2-NEXT: movl %esi, %ecx
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rax
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_c3_load_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: # kill: def $esi killed $esi def $rsi
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shrxq %rsi, (%rdi), %rax
; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%sh_prom = zext i8 %numlowbits to i64
@@ -756,17 +759,17 @@ define i64 @clear_lowbits64_c4_commutati
;
; X64-NOBMI2-LABEL: clear_lowbits64_c4_commutative:
; X64-NOBMI2: # %bb.0:
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: movl %esi, %ecx
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: movq %rsi, %rcx
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_c4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movq $-1, %rax
+; X64-BMI2-NEXT: shrxq %rsi, %rdi, %rax
; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%mask = shl i64 -1, %numlowbits
%masked = and i64 %val, %mask ; swapped order
@@ -782,20 +785,20 @@ define i64 @clear_lowbits64_c4_commutati
define i8 @clear_lowbits8_ic0(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic0:
; X86: # %bb.0:
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
; X86-NEXT: movb $8, %cl
; X86-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_ic0:
; X64: # %bb.0:
; X64-NEXT: movb $8, %cl
; X64-NEXT: subb %sil, %cl
-; X64-NEXT: movb $-1, %al
-; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%numhighbits = sub i8 8, %numlowbits
%mask = shl i8 -1, %numhighbits
@@ -806,21 +809,21 @@ define i8 @clear_lowbits8_ic0(i8 %val, i
define i8 @clear_lowbits8_ic2_load(i8* %w, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic2_load:
; X86: # %bb.0:
-; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movb (%eax), %al
; X86-NEXT: movb $8, %cl
; X86-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb (%edx), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_ic2_load:
; X64: # %bb.0:
+; X64-NEXT: movb (%rdi), %al
; X64-NEXT: movb $8, %cl
; X64-NEXT: subb %sil, %cl
-; X64-NEXT: movb $-1, %al
+; X64-NEXT: shrb %cl, %al
; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb (%rdi), %al
; X64-NEXT: retq
%val = load i8, i8* %w
%numhighbits = sub i8 8, %numlowbits
@@ -832,20 +835,20 @@ define i8 @clear_lowbits8_ic2_load(i8* %
define i8 @clear_lowbits8_ic4_commutative(i8 %val, i8 %numlowbits) nounwind {
; X86-LABEL: clear_lowbits8_ic4_commutative:
; X86: # %bb.0:
+; X86-NEXT: movb {{[0-9]+}}(%esp), %al
; X86-NEXT: movb $8, %cl
; X86-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NEXT: movb $-1, %al
+; X86-NEXT: shrb %cl, %al
; X86-NEXT: shlb %cl, %al
-; X86-NEXT: andb {{[0-9]+}}(%esp), %al
; X86-NEXT: retl
;
; X64-LABEL: clear_lowbits8_ic4_commutative:
; X64: # %bb.0:
; X64-NEXT: movb $8, %cl
; X64-NEXT: subb %sil, %cl
-; X64-NEXT: movb $-1, %al
-; X64-NEXT: shlb %cl, %al
-; X64-NEXT: andb %dil, %al
+; X64-NEXT: shrb %cl, %dil
+; X64-NEXT: shlb %cl, %dil
+; X64-NEXT: movl %edi, %eax
; X64-NEXT: retq
%numhighbits = sub i8 8, %numlowbits
%mask = shl i8 -1, %numhighbits
@@ -858,43 +861,43 @@ define i8 @clear_lowbits8_ic4_commutativ
define i16 @clear_lowbits16_ic0(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic0:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movw $16, %cx
; X86-NOBMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $cx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic0:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movw $16, %ax
-; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %ax
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movw $16, %cx
+; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic0:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movzwl %di, %eax
; X64-NOBMI2-NEXT: movl $16, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic0:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $16, %eax
-; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
-; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: movl $16, %ecx
+; X64-BMI2-NEXT: subl %esi, %ecx
+; X64-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X64-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i16 16, %numlowbits
@@ -906,41 +909,41 @@ define i16 @clear_lowbits16_ic0(i16 %val
define i16 @clear_lowbits16_ic1_indexzext(i16 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movb $16, %cl
; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movb $16, %al
-; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb $16, %cl
+; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movzwl %di, %eax
; X64-NOBMI2-NEXT: movb $16, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic1_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movb $16, %al
-; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movl $-1, %ecx
-; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: movb $16, %cl
+; X64-BMI2-NEXT: subb %sil, %cl
+; X64-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X64-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i8 16, %numlowbits
@@ -953,45 +956,45 @@ define i16 @clear_lowbits16_ic1_indexzex
define i16 @clear_lowbits16_ic2_load(i16* %w, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
; X86-NOBMI2-NEXT: movw $16, %cx
; X86-NOBMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $cx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movw $16, %ax
-; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %ax
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movzwl (%eax), %eax
+; X86-BMI2-NEXT: movw $16, %cx
+; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic2_load:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
; X64-NOBMI2-NEXT: movl $16, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic2_load:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $16, %eax
-; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
-; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: movl $16, %ecx
+; X64-BMI2-NEXT: subl %esi, %ecx
+; X64-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X64-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -1004,43 +1007,43 @@ define i16 @clear_lowbits16_ic2_load(i16
define i16 @clear_lowbits16_ic3_load_indexzext(i16* %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movzwl (%eax), %eax
; X86-NOBMI2-NEXT: movb $16, %cl
; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw (%edx), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb $16, %al
-; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andw (%ecx), %ax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movzwl (%eax), %eax
+; X86-BMI2-NEXT: movb $16, %cl
+; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movzwl (%rdi), %eax
; X64-NOBMI2-NEXT: movb $16, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andw (%rdi), %ax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic3_load_indexzext:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movb $16, %al
-; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movl $-1, %ecx
-; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andw (%rdi), %ax
+; X64-BMI2-NEXT: movzwl (%rdi), %eax
+; X64-BMI2-NEXT: movb $16, %cl
+; X64-BMI2-NEXT: subb %sil, %cl
+; X64-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X64-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%val = load i16, i16* %w
@@ -1054,43 +1057,43 @@ define i16 @clear_lowbits16_ic3_load_ind
define i16 @clear_lowbits16_ic4_commutative(i16 %val, i16 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits16_ic4_commutative:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movw $16, %cx
; X86-NOBMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $cx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
; X86-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits16_ic4_commutative:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movw $16, %ax
-; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %ax
-; X86-BMI2-NEXT: movl $-1, %ecx
-; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andw {{[0-9]+}}(%esp), %ax
+; X86-BMI2-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movw $16, %cx
+; X86-BMI2-NEXT: subw {{[0-9]+}}(%esp), %cx
+; X86-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits16_ic4_commutative:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movzwl %di, %eax
; X64-NOBMI2-NEXT: movl $16, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
; X64-NOBMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits16_ic4_commutative:
; X64-BMI2: # %bb.0:
-; X64-BMI2-NEXT: movl $16, %eax
-; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
-; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
+; X64-BMI2-NEXT: movzwl %di, %eax
+; X64-BMI2-NEXT: movl $16, %ecx
+; X64-BMI2-NEXT: subl %esi, %ecx
+; X64-BMI2-NEXT: shrxl %ecx, %eax, %eax
+; X64-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X64-BMI2-NEXT: # kill: def $ax killed $ax killed $eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i16 16, %numlowbits
@@ -1104,40 +1107,38 @@ define i16 @clear_lowbits16_ic4_commutat
define i32 @clear_lowbits32_ic0(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic0:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movl $32, %ecx
; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic0:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movl $32, %eax
; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %eax
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic0:
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movl $32, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic0:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $32, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
+; X64-BMI2-NEXT: shrxl %eax, %edi, %ecx
; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i32 32, %numlowbits
%mask = shl i32 -1, %numhighbits
@@ -1148,38 +1149,36 @@ define i32 @clear_lowbits32_ic0(i32 %val
define i32 @clear_lowbits32_ic1_indexzext(i32 %val, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movb $32, %cl
; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movb $32, %al
; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movb $32, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movl $-1, %eax
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic1_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movb $32, %al
; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movl $-1, %ecx
+; X64-BMI2-NEXT: shrxl %eax, %edi, %ecx
; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i8 32, %numlowbits
%sh_prom = zext i8 %numhighbits to i32
@@ -1191,42 +1190,40 @@ define i32 @clear_lowbits32_ic1_indexzex
define i32 @clear_lowbits32_ic2_load(i32* %w, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic2_load:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
; X86-NOBMI2-NEXT: movl $32, %ecx
; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic2_load:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movl $32, %eax
-; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %eax
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movl $32, %ecx
+; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI2-NEXT: shrxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic2_load:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
; X64-NOBMI2-NEXT: movl $32, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic2_load:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $32, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
+; X64-BMI2-NEXT: shrxl %eax, (%rdi), %ecx
; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%numhighbits = sub i32 32, %numlowbits
@@ -1238,40 +1235,38 @@ define i32 @clear_lowbits32_ic2_load(i32
define i32 @clear_lowbits32_ic3_load_indexzext(i32* %w, i8 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X86-NOBMI2: # %bb.0:
-; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI2-NEXT: movl (%eax), %eax
; X86-NOBMI2-NEXT: movb $32, %cl
; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl (%edx), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X86-BMI2: # %bb.0:
-; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI2-NEXT: movb $32, %al
-; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al
-; X86-BMI2-NEXT: movl $-1, %edx
-; X86-BMI2-NEXT: shlxl %eax, %edx, %eax
-; X86-BMI2-NEXT: andl (%ecx), %eax
+; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-BMI2-NEXT: movb $32, %cl
+; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl
+; X86-BMI2-NEXT: shrxl %ecx, (%eax), %eax
+; X86-BMI2-NEXT: shlxl %ecx, %eax, %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movl (%rdi), %eax
; X64-NOBMI2-NEXT: movb $32, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %eax
; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl (%rdi), %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic3_load_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movb $32, %al
; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movl $-1, %ecx
+; X64-BMI2-NEXT: shrxl %eax, (%rdi), %ecx
; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl (%rdi), %eax
; X64-BMI2-NEXT: retq
%val = load i32, i32* %w
%numhighbits = sub i8 32, %numlowbits
@@ -1284,40 +1279,38 @@ define i32 @clear_lowbits32_ic3_load_ind
define i32 @clear_lowbits32_ic4_commutative(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI2-LABEL: clear_lowbits32_ic4_commutative:
; X86-NOBMI2: # %bb.0:
+; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: movl $32, %ecx
; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI2-NEXT: movl $-1, %eax
+; X86-NOBMI2-NEXT: shrl %cl, %eax
; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI2-NEXT: shll %cl, %eax
-; X86-NOBMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI2-NEXT: retl
;
; X86-BMI2-LABEL: clear_lowbits32_ic4_commutative:
; X86-BMI2: # %bb.0:
; X86-BMI2-NEXT: movl $32, %eax
; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %eax
-; X86-BMI2-NEXT: movl $-1, %ecx
+; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx
; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X86-BMI2-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-BMI2-NEXT: retl
;
; X64-NOBMI2-LABEL: clear_lowbits32_ic4_commutative:
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movl $32, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movl $-1, %eax
+; X64-NOBMI2-NEXT: shrl %cl, %edi
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI2-NEXT: shll %cl, %eax
-; X64-NOBMI2-NEXT: andl %edi, %eax
+; X64-NOBMI2-NEXT: shll %cl, %edi
+; X64-NOBMI2-NEXT: movl %edi, %eax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits32_ic4_commutative:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $32, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movl $-1, %ecx
+; X64-BMI2-NEXT: shrxl %eax, %edi, %ecx
; X64-BMI2-NEXT: shlxl %eax, %ecx, %eax
-; X64-BMI2-NEXT: andl %edi, %eax
; X64-BMI2-NEXT: retq
%numhighbits = sub i32 32, %numlowbits
%mask = shl i32 -1, %numhighbits
@@ -1367,19 +1360,18 @@ define i64 @clear_lowbits64_ic0(i64 %val
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movl $64, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movq $-1, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic0:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $64, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movq $-1, %rcx
+; X64-BMI2-NEXT: shrxq %rax, %rdi, %rcx
; X64-BMI2-NEXT: shlxq %rax, %rcx, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%numhighbits = sub i64 64, %numlowbits
%mask = shl i64 -1, %numhighbits
@@ -1427,18 +1419,17 @@ define i64 @clear_lowbits64_ic1_indexzex
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movb $64, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movq $-1, %rax
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic1_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movb $64, %al
; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movq $-1, %rcx
+; X64-BMI2-NEXT: shrxq %rax, %rdi, %rcx
; X64-BMI2-NEXT: shlxq %rax, %rcx, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%numhighbits = sub i8 64, %numlowbits
%sh_prom = zext i8 %numhighbits to i64
@@ -1491,21 +1482,20 @@ define i64 @clear_lowbits64_ic2_load(i64
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic2_load:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
; X64-NOBMI2-NEXT: movl $64, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movq $-1, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rax
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic2_load:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $64, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movq $-1, %rcx
+; X64-BMI2-NEXT: shrxq %rax, (%rdi), %rcx
; X64-BMI2-NEXT: shlxq %rax, %rcx, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%numhighbits = sub i64 64, %numlowbits
@@ -1558,20 +1548,19 @@ define i64 @clear_lowbits64_ic3_load_ind
;
; X64-NOBMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X64-NOBMI2: # %bb.0:
+; X64-NOBMI2-NEXT: movq (%rdi), %rax
; X64-NOBMI2-NEXT: movb $64, %cl
; X64-NOBMI2-NEXT: subb %sil, %cl
-; X64-NOBMI2-NEXT: movq $-1, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rax
; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq (%rdi), %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic3_load_indexzext:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movb $64, %al
; X64-BMI2-NEXT: subb %sil, %al
-; X64-BMI2-NEXT: movq $-1, %rcx
+; X64-BMI2-NEXT: shrxq %rax, (%rdi), %rcx
; X64-BMI2-NEXT: shlxq %rax, %rcx, %rax
-; X64-BMI2-NEXT: andq (%rdi), %rax
; X64-BMI2-NEXT: retq
%val = load i64, i64* %w
%numhighbits = sub i8 64, %numlowbits
@@ -1621,19 +1610,18 @@ define i64 @clear_lowbits64_ic4_commutat
; X64-NOBMI2: # %bb.0:
; X64-NOBMI2-NEXT: movl $64, %ecx
; X64-NOBMI2-NEXT: subl %esi, %ecx
-; X64-NOBMI2-NEXT: movq $-1, %rax
+; X64-NOBMI2-NEXT: shrq %cl, %rdi
; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI2-NEXT: shlq %cl, %rax
-; X64-NOBMI2-NEXT: andq %rdi, %rax
+; X64-NOBMI2-NEXT: shlq %cl, %rdi
+; X64-NOBMI2-NEXT: movq %rdi, %rax
; X64-NOBMI2-NEXT: retq
;
; X64-BMI2-LABEL: clear_lowbits64_ic4_commutative:
; X64-BMI2: # %bb.0:
; X64-BMI2-NEXT: movl $64, %eax
; X64-BMI2-NEXT: subl %esi, %eax
-; X64-BMI2-NEXT: movq $-1, %rcx
+; X64-BMI2-NEXT: shrxq %rax, %rdi, %rcx
; X64-BMI2-NEXT: shlxq %rax, %rcx, %rax
-; X64-BMI2-NEXT: andq %rdi, %rax
; X64-BMI2-NEXT: retq
%numhighbits = sub i64 64, %numlowbits
%mask = shl i64 -1, %numhighbits
Modified: llvm/trunk/test/CodeGen/X86/extract-lowbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/extract-lowbits.ll?rev=336585&r1=336584&r2=336585&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/extract-lowbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/extract-lowbits.ll Mon Jul 9 12:06:42 2018
@@ -1015,12 +1015,12 @@ define i64 @bzhi64_b4_commutative(i64 %v
define i32 @bzhi32_c0(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bzhi32_c0:
; X86-NOBMI: # %bb.0:
+; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: movl $32, %ecx
; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI-NEXT: movl $-1, %eax
+; X86-NOBMI-NEXT: shll %cl, %eax
; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT: shrl %cl, %eax
-; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: retl
;
; X86-BMI1BMI2-LABEL: bzhi32_c0:
@@ -1033,10 +1033,10 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movl $32, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movl $-1, %eax
+; X64-NOBMI-NEXT: shll %cl, %edi
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI-NEXT: shrl %cl, %eax
-; X64-NOBMI-NEXT: andl %edi, %eax
+; X64-NOBMI-NEXT: shrl %cl, %edi
+; X64-NOBMI-NEXT: movl %edi, %eax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi32_c0:
@@ -1052,11 +1052,11 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
define i32 @bzhi32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bzhi32_c1_indexzext:
; X86-NOBMI: # %bb.0:
+; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: movb $32, %cl
; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI-NEXT: movl $-1, %eax
+; X86-NOBMI-NEXT: shll %cl, %eax
; X86-NOBMI-NEXT: shrl %cl, %eax
-; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: retl
;
; X86-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
@@ -1069,9 +1069,9 @@ define i32 @bzhi32_c1_indexzext(i32 %val
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movb $32, %cl
; X64-NOBMI-NEXT: subb %sil, %cl
-; X64-NOBMI-NEXT: movl $-1, %eax
-; X64-NOBMI-NEXT: shrl %cl, %eax
-; X64-NOBMI-NEXT: andl %edi, %eax
+; X64-NOBMI-NEXT: shll %cl, %edi
+; X64-NOBMI-NEXT: shrl %cl, %edi
+; X64-NOBMI-NEXT: movl %edi, %eax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
@@ -1088,13 +1088,13 @@ define i32 @bzhi32_c1_indexzext(i32 %val
define i32 @bzhi32_c2_load(i32* %w, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bzhi32_c2_load:
; X86-NOBMI: # %bb.0:
-; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT: movl (%eax), %eax
; X86-NOBMI-NEXT: movl $32, %ecx
; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI-NEXT: movl $-1, %eax
+; X86-NOBMI-NEXT: shll %cl, %eax
; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT: shrl %cl, %eax
-; X86-NOBMI-NEXT: andl (%edx), %eax
; X86-NOBMI-NEXT: retl
;
; X86-BMI1BMI2-LABEL: bzhi32_c2_load:
@@ -1106,12 +1106,12 @@ define i32 @bzhi32_c2_load(i32* %w, i32
;
; X64-NOBMI-LABEL: bzhi32_c2_load:
; X64-NOBMI: # %bb.0:
+; X64-NOBMI-NEXT: movl (%rdi), %eax
; X64-NOBMI-NEXT: movl $32, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movl $-1, %eax
+; X64-NOBMI-NEXT: shll %cl, %eax
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT: shrl %cl, %eax
-; X64-NOBMI-NEXT: andl (%rdi), %eax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi32_c2_load:
@@ -1128,12 +1128,12 @@ define i32 @bzhi32_c2_load(i32* %w, i32
define i32 @bzhi32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bzhi32_c3_load_indexzext:
; X86-NOBMI: # %bb.0:
-; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT: movl (%eax), %eax
; X86-NOBMI-NEXT: movb $32, %cl
; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl
-; X86-NOBMI-NEXT: movl $-1, %eax
+; X86-NOBMI-NEXT: shll %cl, %eax
; X86-NOBMI-NEXT: shrl %cl, %eax
-; X86-NOBMI-NEXT: andl (%edx), %eax
; X86-NOBMI-NEXT: retl
;
; X86-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
@@ -1145,11 +1145,11 @@ define i32 @bzhi32_c3_load_indexzext(i32
;
; X64-NOBMI-LABEL: bzhi32_c3_load_indexzext:
; X64-NOBMI: # %bb.0:
+; X64-NOBMI-NEXT: movl (%rdi), %eax
; X64-NOBMI-NEXT: movb $32, %cl
; X64-NOBMI-NEXT: subb %sil, %cl
-; X64-NOBMI-NEXT: movl $-1, %eax
+; X64-NOBMI-NEXT: shll %cl, %eax
; X64-NOBMI-NEXT: shrl %cl, %eax
-; X64-NOBMI-NEXT: andl (%rdi), %eax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
@@ -1167,12 +1167,12 @@ define i32 @bzhi32_c3_load_indexzext(i32
define i32 @bzhi32_c4_commutative(i32 %val, i32 %numlowbits) nounwind {
; X86-NOBMI-LABEL: bzhi32_c4_commutative:
; X86-NOBMI: # %bb.0:
+; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: movl $32, %ecx
; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx
-; X86-NOBMI-NEXT: movl $-1, %eax
+; X86-NOBMI-NEXT: shll %cl, %eax
; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
; X86-NOBMI-NEXT: shrl %cl, %eax
-; X86-NOBMI-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NOBMI-NEXT: retl
;
; X86-BMI1BMI2-LABEL: bzhi32_c4_commutative:
@@ -1185,10 +1185,10 @@ define i32 @bzhi32_c4_commutative(i32 %v
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movl $32, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movl $-1, %eax
+; X64-NOBMI-NEXT: shll %cl, %edi
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI-NEXT: shrl %cl, %eax
-; X64-NOBMI-NEXT: andl %edi, %eax
+; X64-NOBMI-NEXT: shrl %cl, %edi
+; X64-NOBMI-NEXT: movl %edi, %eax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi32_c4_commutative:
@@ -1243,10 +1243,10 @@ define i64 @bzhi64_c0(i64 %val, i64 %num
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movl $64, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movq $-1, %rax
+; X64-NOBMI-NEXT: shlq %cl, %rdi
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI-NEXT: shrq %cl, %rax
-; X64-NOBMI-NEXT: andq %rdi, %rax
+; X64-NOBMI-NEXT: shrq %cl, %rdi
+; X64-NOBMI-NEXT: movq %rdi, %rax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi64_c0:
@@ -1299,9 +1299,9 @@ define i64 @bzhi64_c1_indexzext(i64 %val
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movb $64, %cl
; X64-NOBMI-NEXT: subb %sil, %cl
-; X64-NOBMI-NEXT: movq $-1, %rax
-; X64-NOBMI-NEXT: shrq %cl, %rax
-; X64-NOBMI-NEXT: andq %rdi, %rax
+; X64-NOBMI-NEXT: shlq %cl, %rdi
+; X64-NOBMI-NEXT: shrq %cl, %rdi
+; X64-NOBMI-NEXT: movq %rdi, %rax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi64_c1_indexzext:
@@ -1360,12 +1360,12 @@ define i64 @bzhi64_c2_load(i64* %w, i64
;
; X64-NOBMI-LABEL: bzhi64_c2_load:
; X64-NOBMI: # %bb.0:
+; X64-NOBMI-NEXT: movq (%rdi), %rax
; X64-NOBMI-NEXT: movl $64, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movq $-1, %rax
+; X64-NOBMI-NEXT: shlq %cl, %rax
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
; X64-NOBMI-NEXT: shrq %cl, %rax
-; X64-NOBMI-NEXT: andq (%rdi), %rax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi64_c2_load:
@@ -1423,11 +1423,11 @@ define i64 @bzhi64_c3_load_indexzext(i64
;
; X64-NOBMI-LABEL: bzhi64_c3_load_indexzext:
; X64-NOBMI: # %bb.0:
+; X64-NOBMI-NEXT: movq (%rdi), %rax
; X64-NOBMI-NEXT: movb $64, %cl
; X64-NOBMI-NEXT: subb %sil, %cl
-; X64-NOBMI-NEXT: movq $-1, %rax
+; X64-NOBMI-NEXT: shlq %cl, %rax
; X64-NOBMI-NEXT: shrq %cl, %rax
-; X64-NOBMI-NEXT: andq (%rdi), %rax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext:
@@ -1483,10 +1483,10 @@ define i64 @bzhi64_c4_commutative(i64 %v
; X64-NOBMI: # %bb.0:
; X64-NOBMI-NEXT: movl $64, %ecx
; X64-NOBMI-NEXT: subl %esi, %ecx
-; X64-NOBMI-NEXT: movq $-1, %rax
+; X64-NOBMI-NEXT: shlq %cl, %rdi
; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx
-; X64-NOBMI-NEXT: shrq %cl, %rax
-; X64-NOBMI-NEXT: andq %rdi, %rax
+; X64-NOBMI-NEXT: shrq %cl, %rdi
+; X64-NOBMI-NEXT: movq %rdi, %rax
; X64-NOBMI-NEXT: retq
;
; X64-BMI1BMI2-LABEL: bzhi64_c4_commutative:
More information about the llvm-commits
mailing list