Index: lib/CodeGen/SelectionDAG/DAGCombiner.cpp =================================================================== --- lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -9779,6 +9779,28 @@ if (SDValue NewVSel = matchVSelectOpSizesWithSetCC(N)) return NewVSel; + // Narrow a suitable binary operation with a constant operand by moving it + // ahead of the truncate. This is limited to pre-legalization because targets + // may prefer a wider type during later combines and invert this transform. + switch (N0.getOpcode()) { + // TODO: Add case for ADD - that will likely require a change in logic here + // or target-specific changes to avoid regressions. + case ISD::SUB: + case ISD::MUL: + case ISD::AND: + case ISD::OR: + case ISD::XOR: + // TODO: This should allow vector constants/types too. + if (!LegalOperations && N0.hasOneUse() && + (isa(N0.getOperand(0)) || + isa(N0.getOperand(1)))) { + SDLoc DL(N); + SDValue NarrowL = DAG.getNode(ISD::TRUNCATE, DL, VT, N0.getOperand(0)); + SDValue NarrowR = DAG.getNode(ISD::TRUNCATE, DL, VT, N0.getOperand(1)); + return DAG.getNode(N0.getOpcode(), DL, VT, NarrowL, NarrowR); + } + } + return SDValue(); } Index: test/CodeGen/AMDGPU/cgp-bitfield-extract.ll =================================================================== --- test/CodeGen/AMDGPU/cgp-bitfield-extract.ll +++ test/CodeGen/AMDGPU/cgp-bitfield-extract.ll @@ -125,11 +125,11 @@ ; GCN: s_cbranch_scc1 ; SI: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80004 -; VI: s_and_b32 s{{[0-9]+}}, [[BFE]], 0xff +; VI: v_mov_b32_e32 v{{[0-9]+}}, 0xff ; GCN: BB2_2: ; SI: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x70004 -; VI: s_and_b32 s{{[0-9]+}}, [[BFE]], 0x7f +; VI: v_mov_b32_e32 v{{[0-9]+}}, 0x7f ; GCN: BB2_3: ; GCN: buffer_store_short Index: test/CodeGen/X86/2008-09-11-CoalescerBug2.ll =================================================================== --- test/CodeGen/X86/2008-09-11-CoalescerBug2.ll +++ test/CodeGen/X86/2008-09-11-CoalescerBug2.ll @@ -17,7 +17,7 @@ ; SOURCE-SCHED-NEXT: setg %cl ; SOURCE-SCHED-NEXT: movb g_73, %dl ; SOURCE-SCHED-NEXT: xorl %eax, %eax -; SOURCE-SCHED-NEXT: subl {{[0-9]+}}(%esp), %eax +; SOURCE-SCHED-NEXT: subb {{[0-9]+}}(%esp), %al ; SOURCE-SCHED-NEXT: testb %dl, %dl ; SOURCE-SCHED-NEXT: jne .LBB0_2 ; SOURCE-SCHED-NEXT: # %bb.1: # %bb11 Index: test/CodeGen/X86/2010-08-04-MaskedSignedCompare.ll =================================================================== --- test/CodeGen/X86/2010-08-04-MaskedSignedCompare.ll +++ test/CodeGen/X86/2010-08-04-MaskedSignedCompare.ll @@ -11,7 +11,7 @@ ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: xorl %eax, %eax ; CHECK-NEXT: cmpq {{.*}}(%rip), %rax -; CHECK-NEXT: sbbl %eax, %eax +; CHECK-NEXT: sbbb %al, %al ; CHECK-NEXT: testb $-106, %al ; CHECK-NEXT: jle .LBB0_1 ; CHECK-NEXT: # %bb.2: # %if.then Index: test/CodeGen/X86/add-sub-nsw-nuw.ll =================================================================== --- test/CodeGen/X86/add-sub-nsw-nuw.ll +++ test/CodeGen/X86/add-sub-nsw-nuw.ll @@ -9,7 +9,7 @@ ; CHECK-LABEL: PR30841: ; CHECK: ## %bb.0: ## %entry ; CHECK-NEXT: movl {{[0-9]+}}(%esp), %eax -; CHECK-NEXT: negl %eax +; CHECK-NEXT: negb %al ; CHECK-NEXT: ## kill: def $al killed $al killed $eax ; CHECK-NEXT: retl entry: Index: test/CodeGen/X86/bool-math.ll =================================================================== --- test/CodeGen/X86/bool-math.ll +++ test/CodeGen/X86/bool-math.ll @@ -33,7 +33,7 @@ ; CHECK-LABEL: sub_zext_cmp_mask_narrower_result: ; CHECK: # %bb.0: ; CHECK-NEXT: movl %edi, %eax -; CHECK-NEXT: andl $1, %eax +; CHECK-NEXT: andb $1, %al ; CHECK-NEXT: orb $46, %al ; CHECK-NEXT: # kill: def $al killed $al killed $eax ; CHECK-NEXT: retq @@ -77,7 +77,7 @@ ; CHECK-LABEL: add_zext_cmp_mask_narrower_result: ; CHECK: # %bb.0: ; CHECK-NEXT: movl %edi, %eax -; CHECK-NEXT: andl $1, %eax +; CHECK-NEXT: andb $1, %al ; CHECK-NEXT: xorb $43, %al ; CHECK-NEXT: # kill: def $al killed $al killed $eax ; CHECK-NEXT: retq @@ -159,7 +159,7 @@ ; CHECK-LABEL: low_bit_select_constants_bigger_true_narrower_result: ; CHECK: # %bb.0: ; CHECK-NEXT: movl %edi, %eax -; CHECK-NEXT: andl $1, %eax +; CHECK-NEXT: andb $1, %al ; CHECK-NEXT: xorb $41, %al ; CHECK-NEXT: # kill: def $al killed $al killed $eax ; CHECK-NEXT: retq Index: test/CodeGen/X86/bt.ll =================================================================== --- test/CodeGen/X86/bt.ll +++ test/CodeGen/X86/bt.ll @@ -1067,7 +1067,7 @@ ; X86: # %bb.0: # %entry ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: btl %eax, %ecx +; X86-NEXT: btl %ecx, %eax ; X86-NEXT: setb %al ; X86-NEXT: retl ; Index: test/CodeGen/X86/clear-lowbits.ll =================================================================== --- test/CodeGen/X86/clear-lowbits.ll +++ test/CodeGen/X86/clear-lowbits.ll @@ -866,10 +866,9 @@ ; 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: movb $16, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; 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: # kill: def $ax killed $ax killed $eax ; X86-NOBMI2-NEXT: retl @@ -877,8 +876,8 @@ ; X86-BMI2-LABEL: clear_lowbits16_ic0: ; X86-BMI2: # %bb.0: ; 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: 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 @@ -887,10 +886,9 @@ ; 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: movb $16, %cl +; X64-NOBMI2-NEXT: subb %sil, %cl ; 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: # kill: def $ax killed $ax killed $eax ; X64-NOBMI2-NEXT: retq @@ -898,8 +896,8 @@ ; X64-BMI2-LABEL: clear_lowbits16_ic0: ; X64-BMI2: # %bb.0: ; X64-BMI2-NEXT: movzwl %di, %eax -; X64-BMI2-NEXT: movl $16, %ecx -; X64-BMI2-NEXT: subl %esi, %ecx +; 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 @@ -962,10 +960,9 @@ ; X86-NOBMI2: # %bb.0: ; 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: movb $16, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; 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: # kill: def $ax killed $ax killed $eax ; X86-NOBMI2-NEXT: retl @@ -974,8 +971,8 @@ ; X86-BMI2: # %bb.0: ; 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: 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 @@ -984,10 +981,9 @@ ; 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: movb $16, %cl +; X64-NOBMI2-NEXT: subb %sil, %cl ; 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: # kill: def $ax killed $ax killed $eax ; X64-NOBMI2-NEXT: retq @@ -995,8 +991,8 @@ ; X64-BMI2-LABEL: clear_lowbits16_ic2_load: ; X64-BMI2: # %bb.0: ; X64-BMI2-NEXT: movzwl (%rdi), %eax -; X64-BMI2-NEXT: movl $16, %ecx -; X64-BMI2-NEXT: subl %esi, %ecx +; 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 @@ -1062,10 +1058,9 @@ ; 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: movb $16, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; 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: # kill: def $ax killed $ax killed $eax ; X86-NOBMI2-NEXT: retl @@ -1073,8 +1068,8 @@ ; X86-BMI2-LABEL: clear_lowbits16_ic4_commutative: ; X86-BMI2: # %bb.0: ; 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: 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 @@ -1083,10 +1078,9 @@ ; 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: movb $16, %cl +; X64-NOBMI2-NEXT: subb %sil, %cl ; 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: # kill: def $ax killed $ax killed $eax ; X64-NOBMI2-NEXT: retq @@ -1094,8 +1088,8 @@ ; X64-BMI2-LABEL: clear_lowbits16_ic4_commutative: ; X64-BMI2: # %bb.0: ; X64-BMI2-NEXT: movzwl %di, %eax -; X64-BMI2-NEXT: movl $16, %ecx -; X64-BMI2-NEXT: subl %esi, %ecx +; 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 @@ -1113,7 +1107,7 @@ ; X86-NOBMI2: # %bb.0: ; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI2-NEXT: xorl %ecx, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: shrl %cl, %eax ; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1122,7 +1116,7 @@ ; X86-BMI2-LABEL: clear_lowbits32_ic0: ; X86-BMI2: # %bb.0: ; X86-BMI2-NEXT: xorl %eax, %eax -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %eax +; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al ; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx ; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax ; X86-BMI2-NEXT: retl @@ -1131,7 +1125,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movl %esi, %ecx ; X64-NOBMI2-NEXT: movl %edi, %eax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrl %cl, %eax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI2-NEXT: shll %cl, %eax @@ -1139,7 +1133,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits32_ic0: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxl %esi, %edi, %eax ; X64-BMI2-NEXT: shlxl %esi, %eax, %eax ; X64-BMI2-NEXT: retq @@ -1197,7 +1191,7 @@ ; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI2-NEXT: movl (%eax), %eax ; X86-NOBMI2-NEXT: xorl %ecx, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: shrl %cl, %eax ; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1207,7 +1201,7 @@ ; X86-BMI2: # %bb.0: ; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI2-NEXT: xorl %ecx, %ecx -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; 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 @@ -1216,7 +1210,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movl %esi, %ecx ; X64-NOBMI2-NEXT: movl (%rdi), %eax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrl %cl, %eax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI2-NEXT: shll %cl, %eax @@ -1224,7 +1218,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits32_ic2_load: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxl %esi, (%rdi), %eax ; X64-BMI2-NEXT: shlxl %esi, %eax, %eax ; X64-BMI2-NEXT: retq @@ -1285,7 +1279,7 @@ ; X86-NOBMI2: # %bb.0: ; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI2-NEXT: xorl %ecx, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: shrl %cl, %eax ; X86-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1294,7 +1288,7 @@ ; X86-BMI2-LABEL: clear_lowbits32_ic4_commutative: ; X86-BMI2: # %bb.0: ; X86-BMI2-NEXT: xorl %eax, %eax -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %eax +; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %al ; X86-BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %ecx ; X86-BMI2-NEXT: shlxl %eax, %ecx, %eax ; X86-BMI2-NEXT: retl @@ -1303,7 +1297,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movl %esi, %ecx ; X64-NOBMI2-NEXT: movl %edi, %eax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrl %cl, %eax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI2-NEXT: shll %cl, %eax @@ -1311,7 +1305,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits32_ic4_commutative: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxl %esi, %edi, %eax ; X64-BMI2-NEXT: shlxl %esi, %eax, %eax ; X64-BMI2-NEXT: retq @@ -1326,8 +1320,8 @@ define i64 @clear_lowbits64_ic0(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI2-LABEL: clear_lowbits64_ic0: ; X86-NOBMI2: # %bb.0: -; X86-NOBMI2-NEXT: movl $64, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: movb $64, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: movl $-1, %edx ; X86-NOBMI2-NEXT: movl $-1, %eax ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1344,8 +1338,8 @@ ; ; X86-BMI2-LABEL: clear_lowbits64_ic0: ; X86-BMI2: # %bb.0: -; X86-BMI2-NEXT: movl $64, %ecx -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI2-NEXT: movb $64, %cl +; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI2-NEXT: movl $-1, %edx ; X86-BMI2-NEXT: shlxl %ecx, %edx, %eax ; X86-BMI2-NEXT: shldl %cl, %edx, %edx @@ -1363,7 +1357,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movq %rsi, %rcx ; X64-NOBMI2-NEXT: movq %rdi, %rax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrq %cl, %rax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI2-NEXT: shlq %cl, %rax @@ -1371,7 +1365,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits64_ic0: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxq %rsi, %rdi, %rax ; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax ; X64-BMI2-NEXT: retq @@ -1446,8 +1440,8 @@ ; X86-NOBMI2: # %bb.0: ; X86-NOBMI2-NEXT: pushl %esi ; X86-NOBMI2-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-NOBMI2-NEXT: movl $64, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: movb $64, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: movl $-1, %edx ; X86-NOBMI2-NEXT: movl $-1, %eax ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1467,8 +1461,8 @@ ; X86-BMI2: # %bb.0: ; X86-BMI2-NEXT: pushl %esi ; X86-BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-BMI2-NEXT: movl $64, %ecx -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI2-NEXT: movb $64, %cl +; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI2-NEXT: movl $-1, %edx ; X86-BMI2-NEXT: shlxl %ecx, %edx, %eax ; X86-BMI2-NEXT: shldl %cl, %edx, %edx @@ -1487,7 +1481,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movq %rsi, %rcx ; X64-NOBMI2-NEXT: movq (%rdi), %rax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrq %cl, %rax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI2-NEXT: shlq %cl, %rax @@ -1495,7 +1489,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits64_ic2_load: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxq %rsi, (%rdi), %rax ; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax ; X64-BMI2-NEXT: retq @@ -1576,8 +1570,8 @@ define i64 @clear_lowbits64_ic4_commutative(i64 %val, i64 %numlowbits) nounwind { ; X86-NOBMI2-LABEL: clear_lowbits64_ic4_commutative: ; X86-NOBMI2: # %bb.0: -; X86-NOBMI2-NEXT: movl $64, %ecx -; X86-NOBMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI2-NEXT: movb $64, %cl +; X86-NOBMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI2-NEXT: movl $-1, %edx ; X86-NOBMI2-NEXT: movl $-1, %eax ; X86-NOBMI2-NEXT: shll %cl, %eax @@ -1594,8 +1588,8 @@ ; ; X86-BMI2-LABEL: clear_lowbits64_ic4_commutative: ; X86-BMI2: # %bb.0: -; X86-BMI2-NEXT: movl $64, %ecx -; X86-BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI2-NEXT: movb $64, %cl +; X86-BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI2-NEXT: movl $-1, %edx ; X86-BMI2-NEXT: shlxl %ecx, %edx, %eax ; X86-BMI2-NEXT: shldl %cl, %edx, %edx @@ -1613,7 +1607,7 @@ ; X64-NOBMI2: # %bb.0: ; X64-NOBMI2-NEXT: movq %rsi, %rcx ; X64-NOBMI2-NEXT: movq %rdi, %rax -; X64-NOBMI2-NEXT: negl %ecx +; X64-NOBMI2-NEXT: negb %cl ; X64-NOBMI2-NEXT: shrq %cl, %rax ; X64-NOBMI2-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI2-NEXT: shlq %cl, %rax @@ -1621,7 +1615,7 @@ ; ; X64-BMI2-LABEL: clear_lowbits64_ic4_commutative: ; X64-BMI2: # %bb.0: -; X64-BMI2-NEXT: negl %esi +; X64-BMI2-NEXT: negb %sil ; X64-BMI2-NEXT: shrxq %rsi, %rdi, %rax ; X64-BMI2-NEXT: shlxq %rsi, %rax, %rax ; X64-BMI2-NEXT: retq Index: test/CodeGen/X86/cmov.ll =================================================================== --- test/CodeGen/X86/cmov.ll +++ test/CodeGen/X86/cmov.ll @@ -81,7 +81,7 @@ ; CHECK-NEXT: movsbl {{.*}}(%rip), %edx ; CHECK-NEXT: movzbl %dl, %ecx ; CHECK-NEXT: shrl $7, %ecx -; CHECK-NEXT: xorl $1, %ecx +; CHECK-NEXT: xorb $1, %cl ; CHECK-NEXT: # kill: def $cl killed $cl killed $ecx ; CHECK-NEXT: sarl %cl, %edx ; CHECK-NEXT: movb {{.*}}(%rip), %al Index: test/CodeGen/X86/extract-bits.ll =================================================================== --- test/CodeGen/X86/extract-bits.ll +++ test/CodeGen/X86/extract-bits.ll @@ -2983,7 +2983,7 @@ ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi ; X86-NOBMI-NEXT: shrl %cl, %edi ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -3005,7 +3005,7 @@ ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -3020,22 +3020,22 @@ ; ; X86-BMI1BMI2-LABEL: bextr32_c0: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: pushl %edi +; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: pushl %eax -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %edi -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax +; X86-BMI1BMI2-NEXT: bzhil %ebx, %esi, %eax ; X86-BMI1BMI2-NEXT: addl $4, %esp ; X86-BMI1BMI2-NEXT: popl %esi -; X86-BMI1BMI2-NEXT: popl %edi +; X86-BMI1BMI2-NEXT: popl %ebx ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bextr32_c0: @@ -3047,7 +3047,7 @@ ; X64-NOBMI-NEXT: movl %edi, %ebx ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %ebx -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl $-1, %ebp ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp @@ -3069,7 +3069,7 @@ ; X64-BMI1NOTBM-NEXT: movl %edi, %ebx ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebx -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebp ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp @@ -3089,8 +3089,8 @@ ; X64-BMI1BMI2-NEXT: pushq %rax ; X64-BMI1BMI2-NEXT: movl %edx, %ebx ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %ebp -; X64-BMI1BMI2-NEXT: movl %edx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -3254,7 +3254,7 @@ ; X86-NOBMI-NEXT: movl (%eax), %edi ; X86-NOBMI-NEXT: shrl %cl, %edi ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -3277,7 +3277,7 @@ ; X86-BMI1NOTBM-NEXT: movl (%eax), %edi ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -3292,23 +3292,23 @@ ; ; X86-BMI1BMI2-LABEL: bextr32_c2_load: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: pushl %edi +; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: pushl %eax -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl -; X86-BMI1BMI2-NEXT: shrxl %ecx, (%eax), %edi -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: shrxl %ecx, (%eax), %esi +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax +; X86-BMI1BMI2-NEXT: bzhil %ebx, %esi, %eax ; X86-BMI1BMI2-NEXT: addl $4, %esp ; X86-BMI1BMI2-NEXT: popl %esi -; X86-BMI1BMI2-NEXT: popl %edi +; X86-BMI1BMI2-NEXT: popl %ebx ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bextr32_c2_load: @@ -3320,7 +3320,7 @@ ; X64-NOBMI-NEXT: movl (%rdi), %ebp ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl $-1, %ebx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebx @@ -3342,7 +3342,7 @@ ; X64-BMI1NOTBM-NEXT: movl (%rdi), %ebp ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebx @@ -3362,8 +3362,8 @@ ; X64-BMI1BMI2-NEXT: pushq %rax ; X64-BMI1BMI2-NEXT: movl %edx, %ebx ; X64-BMI1BMI2-NEXT: shrxl %esi, (%rdi), %ebp -; X64-BMI1BMI2-NEXT: movl %edx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -3531,7 +3531,7 @@ ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi ; X86-NOBMI-NEXT: shrl %cl, %edi ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -3553,7 +3553,7 @@ ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -3568,22 +3568,22 @@ ; ; X86-BMI1BMI2-LABEL: bextr32_c4_commutative: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: pushl %edi +; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: pushl %eax -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %edi -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: shrxl %eax, {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax +; X86-BMI1BMI2-NEXT: bzhil %ebx, %esi, %eax ; X86-BMI1BMI2-NEXT: addl $4, %esp ; X86-BMI1BMI2-NEXT: popl %esi -; X86-BMI1BMI2-NEXT: popl %edi +; X86-BMI1BMI2-NEXT: popl %ebx ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bextr32_c4_commutative: @@ -3595,7 +3595,7 @@ ; X64-NOBMI-NEXT: movl %edi, %ebx ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %ebx -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl $-1, %ebp ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp @@ -3617,7 +3617,7 @@ ; X64-BMI1NOTBM-NEXT: movl %edi, %ebx ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebx -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebp ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp @@ -3637,8 +3637,8 @@ ; X64-BMI1BMI2-NEXT: pushq %rax ; X64-BMI1BMI2-NEXT: movl %edx, %ebx ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %ebp -; X64-BMI1BMI2-NEXT: movl %edx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -3667,7 +3667,7 @@ ; X86-NOBMI-NEXT: movl %ebx, %ecx ; X86-NOBMI-NEXT: shrl %cl, %edi ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -3694,7 +3694,7 @@ ; X86-BMI1NOTBM-NEXT: movl %ebx, %ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -3716,16 +3716,16 @@ ; X86-BMI1BMI2-NEXT: pushl %edi ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: subl $16, %esp -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edi -; X86-BMI1BMI2-NEXT: shrxl %edi, {{[0-9]+}}(%esp), %ebx -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: shrxl %edi, {{[0-9]+}}(%esp), %esi +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, %ebx, %esi +; X86-BMI1BMI2-NEXT: bzhil %ebx, %esi, %esi ; X86-BMI1BMI2-NEXT: movl %edi, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 ; X86-BMI1BMI2-NEXT: movl %esi, %eax @@ -3744,7 +3744,7 @@ ; X64-NOBMI-NEXT: movl %edi, %ebp ; X64-NOBMI-NEXT: movl %r14d, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl $-1, %ebx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebx @@ -3768,7 +3768,7 @@ ; X64-BMI1NOTBM-NEXT: movl %edi, %ebp ; X64-BMI1NOTBM-NEXT: movl %r14d, %ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebx @@ -3791,8 +3791,8 @@ ; X64-BMI1BMI2-NEXT: movl %edx, %ebx ; X64-BMI1BMI2-NEXT: movl %esi, %ebp ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %r14d -; X64-BMI1BMI2-NEXT: movl %edx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -3835,8 +3835,8 @@ ; X86-NOBMI-NEXT: movl %edi, %esi ; X86-NOBMI-NEXT: xorl %edi, %edi ; X86-NOBMI-NEXT: .LBB32_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %ebp ; X86-NOBMI-NEXT: movl $-1, %ebx ; X86-NOBMI-NEXT: shrl %cl, %ebx @@ -3882,8 +3882,8 @@ ; X86-BMI1NOTBM-NEXT: movl %edi, %esi ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi ; X86-BMI1NOTBM-NEXT: .LBB32_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %ebp ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebx @@ -3928,8 +3928,8 @@ ; X86-BMI1BMI2-NEXT: movl %edi, %esi ; X86-BMI1BMI2-NEXT: xorl %edi, %edi ; X86-BMI1BMI2-NEXT: .LBB32_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %ebx ; X86-BMI1BMI2-NEXT: shrxl %ecx, %ebx, %ebp ; X86-BMI1BMI2-NEXT: shrdl %cl, %ebx, %ebx @@ -3964,7 +3964,7 @@ ; X64-NOBMI-NEXT: movq %rdi, %r14 ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %r14 -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -3986,7 +3986,7 @@ ; X64-BMI1NOTBM-NEXT: movq %rdi, %r14 ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %r14 -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -4007,7 +4007,7 @@ ; X64-BMI1BMI2-NEXT: movq %rdx, %rbx ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %r14 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -4257,8 +4257,8 @@ ; X86-NOBMI-NEXT: movl %edi, %esi ; X86-NOBMI-NEXT: xorl %edi, %edi ; X86-NOBMI-NEXT: .LBB34_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %ebp ; X86-NOBMI-NEXT: movl $-1, %ebx ; X86-NOBMI-NEXT: shrl %cl, %ebx @@ -4305,8 +4305,8 @@ ; X86-BMI1NOTBM-NEXT: movl %edi, %esi ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi ; X86-BMI1NOTBM-NEXT: .LBB34_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %ebp ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebx @@ -4352,8 +4352,8 @@ ; X86-BMI1BMI2-NEXT: movl %edi, %esi ; X86-BMI1BMI2-NEXT: xorl %edi, %edi ; X86-BMI1BMI2-NEXT: .LBB34_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %ebx ; X86-BMI1BMI2-NEXT: shrxl %ecx, %ebx, %ebp ; X86-BMI1BMI2-NEXT: shrdl %cl, %ebx, %ebx @@ -4388,7 +4388,7 @@ ; X64-NOBMI-NEXT: movq (%rdi), %r14 ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %r14 -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -4410,7 +4410,7 @@ ; X64-BMI1NOTBM-NEXT: movq (%rdi), %r14 ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %r14 -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -4431,7 +4431,7 @@ ; X64-BMI1BMI2-NEXT: movq %rdx, %rbx ; X64-BMI1BMI2-NEXT: shrxq %rsi, (%rdi), %r14 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -4685,8 +4685,8 @@ ; X86-NOBMI-NEXT: movl %edi, %esi ; X86-NOBMI-NEXT: xorl %edi, %edi ; X86-NOBMI-NEXT: .LBB36_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %ebp ; X86-NOBMI-NEXT: movl $-1, %ebx ; X86-NOBMI-NEXT: shrl %cl, %ebx @@ -4732,8 +4732,8 @@ ; X86-BMI1NOTBM-NEXT: movl %edi, %esi ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi ; X86-BMI1NOTBM-NEXT: .LBB36_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %ebp ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebx @@ -4778,8 +4778,8 @@ ; X86-BMI1BMI2-NEXT: movl %edi, %esi ; X86-BMI1BMI2-NEXT: xorl %edi, %edi ; X86-BMI1BMI2-NEXT: .LBB36_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %ebx ; X86-BMI1BMI2-NEXT: shrxl %ecx, %ebx, %ebp ; X86-BMI1BMI2-NEXT: shrdl %cl, %ebx, %ebx @@ -4814,7 +4814,7 @@ ; X64-NOBMI-NEXT: movq %rdi, %r14 ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %r14 -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -4836,7 +4836,7 @@ ; X64-BMI1NOTBM-NEXT: movq %rdi, %r14 ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %r14 -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -4857,7 +4857,7 @@ ; X64-BMI1BMI2-NEXT: movq %rdx, %rbx ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %r14 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -4894,8 +4894,8 @@ ; X86-NOBMI-NEXT: movl %edi, %esi ; X86-NOBMI-NEXT: xorl %edi, %edi ; X86-NOBMI-NEXT: .LBB37_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %ebx ; X86-NOBMI-NEXT: movl $-1, %ebp ; X86-NOBMI-NEXT: shrl %cl, %ebp @@ -4946,8 +4946,8 @@ ; X86-BMI1NOTBM-NEXT: movl %edi, %esi ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi ; X86-BMI1NOTBM-NEXT: .LBB37_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx ; X86-BMI1NOTBM-NEXT: movl $-1, %ebp ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebp @@ -4997,8 +4997,8 @@ ; X86-BMI1BMI2-NEXT: movl %edi, %esi ; X86-BMI1BMI2-NEXT: xorl %edi, %edi ; X86-BMI1BMI2-NEXT: .LBB37_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %ebp ; X86-BMI1BMI2-NEXT: shrxl %ecx, %ebp, %ebx ; X86-BMI1BMI2-NEXT: shrdl %cl, %ebp, %ebp @@ -5038,7 +5038,7 @@ ; X64-NOBMI-NEXT: movq %rdi, %r15 ; X64-NOBMI-NEXT: movl %r14d, %ecx ; X64-NOBMI-NEXT: shrq %cl, %r15 -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -5062,7 +5062,7 @@ ; X64-BMI1NOTBM-NEXT: movq %rdi, %r15 ; X64-BMI1NOTBM-NEXT: movl %r14d, %ecx ; X64-BMI1NOTBM-NEXT: shrq %cl, %r15 -; X64-BMI1NOTBM-NEXT: negl %edx +; X64-BMI1NOTBM-NEXT: negb %dl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -5086,7 +5086,7 @@ ; X64-BMI1BMI2-NEXT: movq %rsi, %r14 ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %r15 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -5118,7 +5118,7 @@ ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shll %cl, %eax ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %eax @@ -5126,16 +5126,16 @@ ; ; X86-BMI1NOTBM-LABEL: bextr32_d0: ; X86-BMI1NOTBM: # %bb.0: -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1NOTBM-NEXT: movzbl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1NOTBM-NEXT: shll $8, %eax -; X86-BMI1NOTBM-NEXT: orl %ecx, %eax -; X86-BMI1NOTBM-NEXT: bextrl %eax, {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movzbl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: orl %eax, %ecx +; X86-BMI1NOTBM-NEXT: bextrl %ecx, {{[0-9]+}}(%esp), %eax ; X86-BMI1NOTBM-NEXT: retl ; ; X86-BMI1BMI2-LABEL: bextr32_d0: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shrxl %ecx, {{[0-9]+}}(%esp), %ecx ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax @@ -5147,7 +5147,7 @@ ; X64-NOBMI-NEXT: movl %edi, %eax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %eax -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shll %cl, %eax ; X64-NOBMI-NEXT: shrl %cl, %eax @@ -5245,7 +5245,7 @@ ; X86-NOBMI-NEXT: movl (%eax), %eax ; X86-NOBMI-NEXT: shrl %cl, %eax ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shll %cl, %eax ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %eax @@ -5254,16 +5254,16 @@ ; X86-BMI1NOTBM-LABEL: bextr32_d2_load: ; X86-BMI1NOTBM: # %bb.0: ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-BMI1NOTBM-NEXT: movzbl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: shll $8, %ecx -; X86-BMI1NOTBM-NEXT: orl %edx, %ecx -; X86-BMI1NOTBM-NEXT: bextrl %ecx, (%eax), %eax +; X86-BMI1NOTBM-NEXT: movzbl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: orl %ecx, %edx +; X86-BMI1NOTBM-NEXT: bextrl %edx, (%eax), %eax ; X86-BMI1NOTBM-NEXT: retl ; ; X86-BMI1BMI2-LABEL: bextr32_d2_load: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %dl ; X86-BMI1BMI2-NEXT: shrxl %edx, (%ecx), %ecx @@ -5276,7 +5276,7 @@ ; X64-NOBMI-NEXT: movl (%rdi), %eax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %eax -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shll %cl, %eax ; X64-NOBMI-NEXT: shrl %cl, %eax @@ -5381,7 +5381,7 @@ ; X86-NOBMI-NEXT: movl %eax, %ecx ; X86-NOBMI-NEXT: shrl %cl, %esi ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shll %cl, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -5396,7 +5396,7 @@ ; X86-BMI1NOTBM: # %bb.0: ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: subl $8, %esp -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1NOTBM-NEXT: shll $8, %ecx ; X86-BMI1NOTBM-NEXT: movzbl %al, %edx @@ -5413,7 +5413,7 @@ ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: subl $8, %esp -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X86-BMI1BMI2-NEXT: shrxl %ecx, {{[0-9]+}}(%esp), %edx ; X86-BMI1BMI2-NEXT: bzhil %eax, %edx, %esi @@ -5430,7 +5430,7 @@ ; X64-NOBMI-NEXT: movl %edi, %ebx ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: shrl %cl, %ebx -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shll %cl, %ebx ; X64-NOBMI-NEXT: shrl %cl, %ebx @@ -5492,8 +5492,8 @@ ; X86-NOBMI-NEXT: movl %eax, %edi ; X86-NOBMI-NEXT: xorl %eax, %eax ; X86-NOBMI-NEXT: .LBB43_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shldl %cl, %edi, %eax ; X86-NOBMI-NEXT: shll %cl, %edi ; X86-NOBMI-NEXT: testb $32, %cl @@ -5540,8 +5540,8 @@ ; X86-BMI1NOTBM-NEXT: movl %eax, %edi ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax ; X86-BMI1NOTBM-NEXT: .LBB43_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %eax ; X86-BMI1NOTBM-NEXT: shll %cl, %edi ; X86-BMI1NOTBM-NEXT: testb $32, %cl @@ -5586,8 +5586,8 @@ ; X86-BMI1BMI2-NEXT: movl %esi, %eax ; X86-BMI1BMI2-NEXT: xorl %esi, %esi ; X86-BMI1BMI2-NEXT: .LBB43_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %esi ; X86-BMI1BMI2-NEXT: shlxl %ecx, %eax, %edi ; X86-BMI1BMI2-NEXT: testb $32, %cl @@ -5617,7 +5617,7 @@ ; X64-NOBMI-NEXT: movq %rdi, %rax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rax -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shlq %cl, %rax ; X64-NOBMI-NEXT: shrq %cl, %rax @@ -5838,8 +5838,8 @@ ; X86-NOBMI-NEXT: movl %eax, %edi ; X86-NOBMI-NEXT: xorl %eax, %eax ; X86-NOBMI-NEXT: .LBB45_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shldl %cl, %edi, %eax ; X86-NOBMI-NEXT: shll %cl, %edi ; X86-NOBMI-NEXT: testb $32, %cl @@ -5887,8 +5887,8 @@ ; X86-BMI1NOTBM-NEXT: movl %eax, %edi ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax ; X86-BMI1NOTBM-NEXT: .LBB45_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %eax ; X86-BMI1NOTBM-NEXT: shll %cl, %edi ; X86-BMI1NOTBM-NEXT: testb $32, %cl @@ -5934,8 +5934,8 @@ ; X86-BMI1BMI2-NEXT: movl %esi, %eax ; X86-BMI1BMI2-NEXT: xorl %esi, %esi ; X86-BMI1BMI2-NEXT: .LBB45_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %esi ; X86-BMI1BMI2-NEXT: shlxl %ecx, %eax, %edi ; X86-BMI1BMI2-NEXT: testb $32, %cl @@ -5965,7 +5965,7 @@ ; X64-NOBMI-NEXT: movq (%rdi), %rax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rax -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shlq %cl, %rax ; X64-NOBMI-NEXT: shrq %cl, %rax @@ -6193,8 +6193,8 @@ ; X86-NOBMI-NEXT: movl %esi, %ebx ; X86-NOBMI-NEXT: xorl %esi, %esi ; X86-NOBMI-NEXT: .LBB47_2: -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shldl %cl, %ebx, %esi ; X86-NOBMI-NEXT: shll %cl, %ebx ; X86-NOBMI-NEXT: testb $32, %cl @@ -6254,8 +6254,8 @@ ; X86-BMI1NOTBM-NEXT: movl %esi, %ebx ; X86-BMI1NOTBM-NEXT: xorl %esi, %esi ; X86-BMI1NOTBM-NEXT: .LBB47_2: -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: shldl %cl, %ebx, %esi ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx ; X86-BMI1NOTBM-NEXT: testb $32, %cl @@ -6312,8 +6312,8 @@ ; X86-BMI1BMI2-NEXT: movl %edx, %edi ; X86-BMI1BMI2-NEXT: xorl %edx, %edx ; X86-BMI1BMI2-NEXT: .LBB47_2: -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edx ; X86-BMI1BMI2-NEXT: shlxl %ecx, %edi, %ebx ; X86-BMI1BMI2-NEXT: testb $32, %cl @@ -6352,7 +6352,7 @@ ; X64-NOBMI-NEXT: movq %rdi, %rbx ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: shrq %cl, %rbx -; X64-NOBMI-NEXT: negl %edx +; X64-NOBMI-NEXT: negb %dl ; X64-NOBMI-NEXT: movl %edx, %ecx ; X64-NOBMI-NEXT: shlq %cl, %rbx ; X64-NOBMI-NEXT: shrq %cl, %rbx Index: test/CodeGen/X86/extract-lowbits.ll =================================================================== --- test/CodeGen/X86/extract-lowbits.ll +++ test/CodeGen/X86/extract-lowbits.ll @@ -1436,7 +1436,7 @@ ; X86-NOBMI-NEXT: pushl %esi ; X86-NOBMI-NEXT: subl $8, %esp ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -1453,7 +1453,7 @@ ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: subl $8, %esp ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -1467,18 +1467,18 @@ ; ; X86-BMI1BMI2-LABEL: bzhi32_c0: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: pushl %esi +; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: subl $8, %esp -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: bzhil %ebx, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: addl $8, %esp -; X86-BMI1BMI2-NEXT: popl %esi +; X86-BMI1BMI2-NEXT: popl %ebx ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_c0: @@ -1488,7 +1488,7 @@ ; X64-NOBMI-NEXT: pushq %rax ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: movl %edi, %ebx -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movl $-1, %ebp ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp @@ -1508,7 +1508,7 @@ ; X64-BMI1NOTBM-NEXT: pushq %rax ; X64-BMI1NOTBM-NEXT: movl %esi, %ecx ; X64-BMI1NOTBM-NEXT: movl %edi, %ebx -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebp ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp @@ -1528,8 +1528,8 @@ ; X64-BMI1BMI2-NEXT: pushq %rax ; X64-BMI1BMI2-NEXT: movl %esi, %ebx ; X64-BMI1BMI2-NEXT: movl %edi, %ebp -; X64-BMI1BMI2-NEXT: movl %esi, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -1668,7 +1668,7 @@ ; X86-NOBMI-NEXT: subl $8, %esp ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %edx ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %edx @@ -1687,7 +1687,7 @@ ; X86-BMI1NOTBM-NEXT: subl $8, %esp ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %edx ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx @@ -1705,9 +1705,10 @@ ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: subl $8, %esp ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %esi -; X86-BMI1BMI2-NEXT: negl %ecx +; X86-BMI1BMI2-NEXT: # kill: def $cl killed $cl killed $ecx def $ecx +; X86-BMI1BMI2-NEXT: negb %cl ; X86-BMI1BMI2-NEXT: movl $-1, %eax ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) @@ -1721,7 +1722,7 @@ ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: pushq %rbx ; X64-NOBMI-NEXT: movl %esi, %ecx -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movl $-1, %eax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %eax @@ -1737,7 +1738,7 @@ ; X64-BMI1NOTBM: # %bb.0: ; X64-BMI1NOTBM-NEXT: pushq %rbx ; X64-BMI1NOTBM-NEXT: movl %esi, %ecx -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movl $-1, %eax ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %eax @@ -1753,7 +1754,8 @@ ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: pushq %rbx ; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %ebx -; X64-BMI1BMI2-NEXT: negl %esi +; X64-BMI1BMI2-NEXT: # kill: def $sil killed $sil killed $esi def $esi +; X64-BMI1BMI2-NEXT: negb %sil ; X64-BMI1BMI2-NEXT: movl $-1, %eax ; X64-BMI1BMI2-NEXT: shrxl %esi, %eax, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -1884,7 +1886,7 @@ ; X86-NOBMI-NEXT: pushl %esi ; X86-NOBMI-NEXT: subl $8, %esp ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %esi @@ -1901,7 +1903,7 @@ ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: subl $8, %esp ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi @@ -1915,18 +1917,18 @@ ; ; X86-BMI1BMI2-LABEL: bzhi32_c4_commutative: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: pushl %esi +; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: subl $8, %esp -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-BMI1BMI2-NEXT: movl %esi, %eax -; X86-BMI1BMI2-NEXT: negl %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl +; X86-BMI1BMI2-NEXT: movl %ebx, %eax +; X86-BMI1BMI2-NEXT: negb %al ; X86-BMI1BMI2-NEXT: movl $-1, %ecx ; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax ; X86-BMI1BMI2-NEXT: movl %eax, (%esp) ; X86-BMI1BMI2-NEXT: calll use32 -; X86-BMI1BMI2-NEXT: bzhil %esi, {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: bzhil %ebx, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: addl $8, %esp -; X86-BMI1BMI2-NEXT: popl %esi +; X86-BMI1BMI2-NEXT: popl %ebx ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_c4_commutative: @@ -1936,7 +1938,7 @@ ; X64-NOBMI-NEXT: pushq %rax ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: movl %edi, %ebx -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movl $-1, %ebp ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %ebp @@ -1956,7 +1958,7 @@ ; X64-BMI1NOTBM-NEXT: pushq %rax ; X64-BMI1NOTBM-NEXT: movl %esi, %ecx ; X64-BMI1NOTBM-NEXT: movl %edi, %ebx -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movl $-1, %ebp ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-BMI1NOTBM-NEXT: shrl %cl, %ebp @@ -1976,8 +1978,8 @@ ; X64-BMI1BMI2-NEXT: pushq %rax ; X64-BMI1BMI2-NEXT: movl %esi, %ebx ; X64-BMI1BMI2-NEXT: movl %edi, %ebp -; X64-BMI1BMI2-NEXT: movl %esi, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: movl %ebx, %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movl $-1, %ecx ; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %edi ; X64-BMI1BMI2-NEXT: callq use32 @@ -2003,8 +2005,8 @@ ; X86-NOBMI-NEXT: pushl %edi ; X86-NOBMI-NEXT: pushl %esi ; X86-NOBMI-NEXT: pushl %eax -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: movl $-1, %edi ; X86-NOBMI-NEXT: shrl %cl, %edi @@ -2034,8 +2036,8 @@ ; X86-BMI1NOTBM-NEXT: pushl %edi ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: pushl %eax -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: movl $-1, %edi ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi @@ -2065,8 +2067,8 @@ ; X86-BMI1BMI2-NEXT: pushl %edi ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: pushl %eax -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %esi ; X86-BMI1BMI2-NEXT: shrxl %ecx, %esi, %edi ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %esi @@ -2097,7 +2099,7 @@ ; X64-NOBMI-NEXT: pushq %rax ; X64-NOBMI-NEXT: movq %rsi, %rcx ; X64-NOBMI-NEXT: movq %rdi, %r14 -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -2117,7 +2119,7 @@ ; X64-BMI1NOTBM-NEXT: pushq %rax ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx ; X64-BMI1NOTBM-NEXT: movq %rdi, %r14 -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -2138,7 +2140,7 @@ ; X64-BMI1BMI2-NEXT: movq %rsi, %rbx ; X64-BMI1BMI2-NEXT: movq %rdi, %r14 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -2318,26 +2320,26 @@ ; X86-NOBMI-NEXT: pushl %ebx ; X86-NOBMI-NEXT: pushl %edi ; X86-NOBMI-NEXT: pushl %esi -; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx -; X86-NOBMI-NEXT: movl $-1, %edx +; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-NOBMI-NEXT: movl $-1, %eax ; X86-NOBMI-NEXT: movl $-1, %ebx ; X86-NOBMI-NEXT: shrl %cl, %ebx -; X86-NOBMI-NEXT: shrdl %cl, %edx, %edx +; X86-NOBMI-NEXT: shrdl %cl, %eax, %eax ; X86-NOBMI-NEXT: testb $32, %cl ; X86-NOBMI-NEXT: je .LBB27_2 ; X86-NOBMI-NEXT: # %bb.1: -; X86-NOBMI-NEXT: movl %ebx, %edx +; X86-NOBMI-NEXT: movl %ebx, %eax ; X86-NOBMI-NEXT: xorl %ebx, %ebx ; X86-NOBMI-NEXT: .LBB27_2: -; X86-NOBMI-NEXT: movl (%eax), %esi -; X86-NOBMI-NEXT: andl %edx, %esi -; X86-NOBMI-NEXT: movl 4(%eax), %edi +; X86-NOBMI-NEXT: movl (%edx), %esi +; X86-NOBMI-NEXT: andl %eax, %esi +; X86-NOBMI-NEXT: movl 4(%edx), %edi ; X86-NOBMI-NEXT: andl %ebx, %edi ; X86-NOBMI-NEXT: subl $8, %esp ; X86-NOBMI-NEXT: pushl %ebx -; X86-NOBMI-NEXT: pushl %edx +; X86-NOBMI-NEXT: pushl %eax ; X86-NOBMI-NEXT: calll use64 ; X86-NOBMI-NEXT: addl $16, %esp ; X86-NOBMI-NEXT: movl %esi, %eax @@ -2352,26 +2354,26 @@ ; X86-BMI1NOTBM-NEXT: pushl %ebx ; X86-BMI1NOTBM-NEXT: pushl %edi ; X86-BMI1NOTBM-NEXT: pushl %esi -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx -; X86-BMI1NOTBM-NEXT: movl $-1, %edx +; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1NOTBM-NEXT: movl $-1, %eax ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebx -; X86-BMI1NOTBM-NEXT: shrdl %cl, %edx, %edx +; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %eax ; X86-BMI1NOTBM-NEXT: testb $32, %cl ; X86-BMI1NOTBM-NEXT: je .LBB27_2 ; X86-BMI1NOTBM-NEXT: # %bb.1: -; X86-BMI1NOTBM-NEXT: movl %ebx, %edx +; X86-BMI1NOTBM-NEXT: movl %ebx, %eax ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx ; X86-BMI1NOTBM-NEXT: .LBB27_2: -; X86-BMI1NOTBM-NEXT: movl (%eax), %esi -; X86-BMI1NOTBM-NEXT: andl %edx, %esi -; X86-BMI1NOTBM-NEXT: movl 4(%eax), %edi +; X86-BMI1NOTBM-NEXT: movl (%edx), %esi +; X86-BMI1NOTBM-NEXT: andl %eax, %esi +; X86-BMI1NOTBM-NEXT: movl 4(%edx), %edi ; X86-BMI1NOTBM-NEXT: andl %ebx, %edi ; X86-BMI1NOTBM-NEXT: subl $8, %esp ; X86-BMI1NOTBM-NEXT: pushl %ebx -; X86-BMI1NOTBM-NEXT: pushl %edx +; X86-BMI1NOTBM-NEXT: pushl %eax ; X86-BMI1NOTBM-NEXT: calll use64 ; X86-BMI1NOTBM-NEXT: addl $16, %esp ; X86-BMI1NOTBM-NEXT: movl %esi, %eax @@ -2386,25 +2388,25 @@ ; X86-BMI1BMI2-NEXT: pushl %ebx ; X86-BMI1BMI2-NEXT: pushl %edi ; X86-BMI1BMI2-NEXT: pushl %esi -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx -; X86-BMI1BMI2-NEXT: movl $-1, %edx -; X86-BMI1BMI2-NEXT: shrxl %ecx, %edx, %ebx -; X86-BMI1BMI2-NEXT: shrdl %cl, %edx, %edx +; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl +; X86-BMI1BMI2-NEXT: movl $-1, %eax +; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %ebx +; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %eax ; X86-BMI1BMI2-NEXT: testb $32, %cl ; X86-BMI1BMI2-NEXT: je .LBB27_2 ; X86-BMI1BMI2-NEXT: # %bb.1: -; X86-BMI1BMI2-NEXT: movl %ebx, %edx +; X86-BMI1BMI2-NEXT: movl %ebx, %eax ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx ; X86-BMI1BMI2-NEXT: .LBB27_2: -; X86-BMI1BMI2-NEXT: movl (%eax), %esi -; X86-BMI1BMI2-NEXT: andl %edx, %esi -; X86-BMI1BMI2-NEXT: movl 4(%eax), %edi +; X86-BMI1BMI2-NEXT: movl (%edx), %esi +; X86-BMI1BMI2-NEXT: andl %eax, %esi +; X86-BMI1BMI2-NEXT: movl 4(%edx), %edi ; X86-BMI1BMI2-NEXT: andl %ebx, %edi ; X86-BMI1BMI2-NEXT: subl $8, %esp ; X86-BMI1BMI2-NEXT: pushl %ebx -; X86-BMI1BMI2-NEXT: pushl %edx +; X86-BMI1BMI2-NEXT: pushl %eax ; X86-BMI1BMI2-NEXT: calll use64 ; X86-BMI1BMI2-NEXT: addl $16, %esp ; X86-BMI1BMI2-NEXT: movl %esi, %eax @@ -2418,7 +2420,7 @@ ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: pushq %rbx ; X64-NOBMI-NEXT: movq %rsi, %rcx -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movq $-1, %rax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rax @@ -2434,7 +2436,7 @@ ; X64-BMI1NOTBM: # %bb.0: ; X64-BMI1NOTBM-NEXT: pushq %rbx ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movq $-1, %rax ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rax @@ -2450,8 +2452,8 @@ ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: pushq %rbx ; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rbx -; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi killed $rsi def $rsi -; X64-BMI1BMI2-NEXT: negl %esi +; X64-BMI1BMI2-NEXT: # kill: def $sil killed $sil killed $rsi def $rsi +; X64-BMI1BMI2-NEXT: negb %sil ; X64-BMI1BMI2-NEXT: movq $-1, %rax ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rax, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -2628,8 +2630,8 @@ ; X86-NOBMI-NEXT: pushl %edi ; X86-NOBMI-NEXT: pushl %esi ; X86-NOBMI-NEXT: pushl %eax -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl $-1, %esi ; X86-NOBMI-NEXT: movl $-1, %edi ; X86-NOBMI-NEXT: shrl %cl, %edi @@ -2659,8 +2661,8 @@ ; X86-BMI1NOTBM-NEXT: pushl %edi ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: pushl %eax -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl $-1, %esi ; X86-BMI1NOTBM-NEXT: movl $-1, %edi ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi @@ -2690,8 +2692,8 @@ ; X86-BMI1BMI2-NEXT: pushl %edi ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: pushl %eax -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: movl $-1, %esi ; X86-BMI1BMI2-NEXT: shrxl %ecx, %esi, %edi ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %esi @@ -2722,7 +2724,7 @@ ; X64-NOBMI-NEXT: pushq %rax ; X64-NOBMI-NEXT: movq %rsi, %rcx ; X64-NOBMI-NEXT: movq %rdi, %r14 -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: movq $-1, %rbx ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rbx @@ -2742,7 +2744,7 @@ ; X64-BMI1NOTBM-NEXT: pushq %rax ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx ; X64-BMI1NOTBM-NEXT: movq %rdi, %r14 -; X64-BMI1NOTBM-NEXT: negl %ecx +; X64-BMI1NOTBM-NEXT: negb %cl ; X64-BMI1NOTBM-NEXT: movq $-1, %rbx ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-BMI1NOTBM-NEXT: shrq %cl, %rbx @@ -2763,7 +2765,7 @@ ; X64-BMI1BMI2-NEXT: movq %rsi, %rbx ; X64-BMI1BMI2-NEXT: movq %rdi, %r14 ; X64-BMI1BMI2-NEXT: movl %ebx, %eax -; X64-BMI1BMI2-NEXT: negl %eax +; X64-BMI1BMI2-NEXT: negb %al ; X64-BMI1BMI2-NEXT: movq $-1, %rcx ; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rdi ; X64-BMI1BMI2-NEXT: callq use64 @@ -2788,7 +2790,7 @@ ; X86-NOBMI: # %bb.0: ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shll %cl, %eax ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %eax @@ -2796,14 +2798,14 @@ ; ; X86-BMI1NOTBM-LABEL: bzhi32_d0: ; X86-BMI1NOTBM: # %bb.0: -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1NOTBM-NEXT: shll $8, %eax ; X86-BMI1NOTBM-NEXT: bextrl %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1NOTBM-NEXT: retl ; ; X86-BMI1BMI2-LABEL: bzhi32_d0: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al ; X86-BMI1BMI2-NEXT: bzhil %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: retl ; @@ -2811,7 +2813,7 @@ ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: movl %edi, %eax -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: shll %cl, %eax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %eax @@ -2890,7 +2892,7 @@ ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: movl (%eax), %eax ; X86-NOBMI-NEXT: xorl %ecx, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: shll %cl, %eax ; X86-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X86-NOBMI-NEXT: shrl %cl, %eax @@ -2899,7 +2901,7 @@ ; X86-BMI1NOTBM-LABEL: bzhi32_d2_load: ; X86-BMI1NOTBM: # %bb.0: ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: shll $8, %ecx ; X86-BMI1NOTBM-NEXT: bextrl %ecx, (%eax), %eax ; X86-BMI1NOTBM-NEXT: retl @@ -2907,15 +2909,15 @@ ; X86-BMI1BMI2-LABEL: bzhi32_d2_load: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-BMI1BMI2-NEXT: bzhil %eax, (%ecx), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_d2_load: ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: movl %esi, %ecx ; X64-NOBMI-NEXT: movl (%rdi), %eax -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: shll %cl, %eax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx ; X64-NOBMI-NEXT: shrl %cl, %eax @@ -3003,8 +3005,8 @@ ; X86-NOBMI-NEXT: pushl %esi ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edx ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl %edx, %esi ; X86-NOBMI-NEXT: shll %cl, %esi ; X86-NOBMI-NEXT: shldl %cl, %edx, %eax @@ -3042,8 +3044,8 @@ ; X86-BMI1NOTBM-NEXT: pushl %esi ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl %edx, %esi ; X86-BMI1NOTBM-NEXT: shll %cl, %esi ; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax @@ -3080,8 +3082,8 @@ ; X86-BMI1BMI2-NEXT: pushl %esi ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %esi ; X86-BMI1BMI2-NEXT: shlxl %ecx, %eax, %edi ; X86-BMI1BMI2-NEXT: xorl %edx, %edx @@ -3110,7 +3112,7 @@ ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: movq %rsi, %rcx ; X64-NOBMI-NEXT: movq %rdi, %rax -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: shlq %cl, %rax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rax @@ -3281,8 +3283,8 @@ ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NOBMI-NEXT: movl (%eax), %edx ; X86-NOBMI-NEXT: movl 4(%eax), %eax -; X86-NOBMI-NEXT: movl $64, %ecx -; X86-NOBMI-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-NOBMI-NEXT: movb $64, %cl +; X86-NOBMI-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-NOBMI-NEXT: movl %edx, %esi ; X86-NOBMI-NEXT: shll %cl, %esi ; X86-NOBMI-NEXT: shldl %cl, %edx, %eax @@ -3321,8 +3323,8 @@ ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1NOTBM-NEXT: movl (%eax), %edx ; X86-BMI1NOTBM-NEXT: movl 4(%eax), %eax -; X86-BMI1NOTBM-NEXT: movl $64, %ecx -; X86-BMI1NOTBM-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1NOTBM-NEXT: movb $64, %cl +; X86-BMI1NOTBM-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1NOTBM-NEXT: movl %edx, %esi ; X86-BMI1NOTBM-NEXT: shll %cl, %esi ; X86-BMI1NOTBM-NEXT: shldl %cl, %edx, %eax @@ -3360,8 +3362,8 @@ ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: movl (%eax), %edx ; X86-BMI1BMI2-NEXT: movl 4(%eax), %esi -; X86-BMI1BMI2-NEXT: movl $64, %ecx -; X86-BMI1BMI2-NEXT: subl {{[0-9]+}}(%esp), %ecx +; X86-BMI1BMI2-NEXT: movb $64, %cl +; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl ; X86-BMI1BMI2-NEXT: shldl %cl, %edx, %esi ; X86-BMI1BMI2-NEXT: shlxl %ecx, %edx, %edi ; X86-BMI1BMI2-NEXT: xorl %edx, %edx @@ -3390,7 +3392,7 @@ ; X64-NOBMI: # %bb.0: ; X64-NOBMI-NEXT: movq %rsi, %rcx ; X64-NOBMI-NEXT: movq (%rdi), %rax -; X64-NOBMI-NEXT: negl %ecx +; X64-NOBMI-NEXT: negb %cl ; X64-NOBMI-NEXT: shlq %cl, %rax ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx ; X64-NOBMI-NEXT: shrq %cl, %rax Index: test/CodeGen/X86/funnel-shift-rot.ll =================================================================== --- test/CodeGen/X86/funnel-shift-rot.ll +++ test/CodeGen/X86/funnel-shift-rot.ll @@ -231,30 +231,25 @@ ; X32-SSE2-NEXT: pushl %ebx ; X32-SSE2-NEXT: pushl %edi ; X32-SSE2-NEXT: pushl %esi +; X32-SSE2-NEXT: movb {{[0-9]+}}(%esp), %cl ; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %esi ; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %edx -; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ebx -; X32-SSE2-NEXT: movl %ebx, %ecx -; X32-SSE2-NEXT: andl $63, %ecx ; X32-SSE2-NEXT: movl %edx, %edi ; X32-SSE2-NEXT: shrl %cl, %edi -; X32-SSE2-NEXT: movl %esi, %ebp -; X32-SSE2-NEXT: shrdl %cl, %edx, %ebp -; X32-SSE2-NEXT: xorl %eax, %eax +; X32-SSE2-NEXT: movl %esi, %ebx +; X32-SSE2-NEXT: shrdl %cl, %edx, %ebx +; X32-SSE2-NEXT: xorl %ebp, %ebp ; X32-SSE2-NEXT: testb $32, %cl -; X32-SSE2-NEXT: cmovnel %edi, %ebp -; X32-SSE2-NEXT: cmovnel %eax, %edi -; X32-SSE2-NEXT: negl %ebx -; X32-SSE2-NEXT: andl $63, %ebx +; X32-SSE2-NEXT: cmovnel %edi, %ebx +; X32-SSE2-NEXT: cmovnel %ebp, %edi +; X32-SSE2-NEXT: negb %cl ; X32-SSE2-NEXT: movl %esi, %eax -; X32-SSE2-NEXT: movl %ebx, %ecx ; X32-SSE2-NEXT: shll %cl, %eax ; X32-SSE2-NEXT: shldl %cl, %esi, %edx -; X32-SSE2-NEXT: testb $32, %bl +; X32-SSE2-NEXT: testb $32, %cl ; X32-SSE2-NEXT: cmovnel %eax, %edx -; X32-SSE2-NEXT: movl $0, %ecx -; X32-SSE2-NEXT: cmovnel %ecx, %eax -; X32-SSE2-NEXT: orl %ebp, %eax +; X32-SSE2-NEXT: cmovnel %ebp, %eax +; X32-SSE2-NEXT: orl %ebx, %eax ; X32-SSE2-NEXT: orl %edi, %edx ; X32-SSE2-NEXT: popl %esi ; X32-SSE2-NEXT: popl %edi Index: test/CodeGen/X86/funnel-shift.ll =================================================================== --- test/CodeGen/X86/funnel-shift.ll +++ test/CodeGen/X86/funnel-shift.ll @@ -78,8 +78,8 @@ ; X32-SSE2-NEXT: cmovnel %ebp, %eax ; X32-SSE2-NEXT: cmovnel %ecx, %ebp ; X32-SSE2-NEXT: xorl %edx, %edx -; X32-SSE2-NEXT: movl $37, %ecx -; X32-SSE2-NEXT: subl %ebx, %ecx +; X32-SSE2-NEXT: movb $37, %cl +; X32-SSE2-NEXT: subb %bl, %cl ; X32-SSE2-NEXT: shrdl %cl, %esi, %edi ; X32-SSE2-NEXT: shrl %cl, %esi ; X32-SSE2-NEXT: testb $32, %cl @@ -264,8 +264,8 @@ ; X32-SSE2-NEXT: calll __umoddi3 ; X32-SSE2-NEXT: addl $16, %esp ; X32-SSE2-NEXT: movl %eax, %ebx -; X32-SSE2-NEXT: movl $37, %ecx -; X32-SSE2-NEXT: subl %eax, %ecx +; X32-SSE2-NEXT: movb $37, %cl +; X32-SSE2-NEXT: subb %bl, %cl ; X32-SSE2-NEXT: movl %ebp, %eax ; X32-SSE2-NEXT: shll %cl, %ebp ; X32-SSE2-NEXT: shldl %cl, %eax, %edi Index: test/CodeGen/X86/pr32284.ll =================================================================== --- test/CodeGen/X86/pr32284.ll +++ test/CodeGen/X86/pr32284.ll @@ -81,9 +81,8 @@ ; 686-NEXT: movzbl c, %eax ; 686-NEXT: xorl %ecx, %ecx ; 686-NEXT: testl %eax, %eax -; 686-NEXT: setne %cl -; 686-NEXT: testb %al, %al ; 686-NEXT: setne {{[0-9]+}}(%esp) +; 686-NEXT: setne %cl ; 686-NEXT: xorl %edx, %edx ; 686-NEXT: cmpl %eax, %ecx ; 686-NEXT: setle %dl Index: test/CodeGen/X86/pr37879.ll =================================================================== --- test/CodeGen/X86/pr37879.ll +++ test/CodeGen/X86/pr37879.ll @@ -6,8 +6,6 @@ ; CHECK: ## %bb.0: ; CHECK-NEXT: movq (%rax), %rax ; CHECK-NEXT: vcvtsi2sdq %rax, %xmm0, %xmm1 -; CHECK-NEXT: ## kill: def $eax killed $eax killed $rax -; CHECK-NEXT: andl $1, %eax ; CHECK-NEXT: kmovd %eax, %k1 ; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; CHECK-NEXT: vmovsd %xmm1, %xmm0, %xmm0 {%k1} Index: test/CodeGen/X86/rot16.ll =================================================================== --- test/CodeGen/X86/rot16.ll +++ test/CodeGen/X86/rot16.ll @@ -15,7 +15,7 @@ ; X64-NEXT: movl %edx, %ecx ; X64-NEXT: movl %edi, %eax ; X64-NEXT: # kill: def $cl killed $cl killed $ecx -; X64-NEXT: shldw %cl, %ax, %ax +; X64-NEXT: rolw %cl, %ax ; X64-NEXT: # kill: def $ax killed $ax killed $eax ; X64-NEXT: retq %t0 = shl i16 %x, %z @@ -62,7 +62,7 @@ ; X64-NEXT: movl %edx, %ecx ; X64-NEXT: movl %edi, %eax ; X64-NEXT: # kill: def $cl killed $cl killed $ecx -; X64-NEXT: shrdw %cl, %ax, %ax +; X64-NEXT: rorw %cl, %ax ; X64-NEXT: # kill: def $ax killed $ax killed $eax ; X64-NEXT: retq %t0 = lshr i16 %x, %z Index: test/CodeGen/X86/rotate.ll =================================================================== --- test/CodeGen/X86/rotate.ll +++ test/CodeGen/X86/rotate.ll @@ -653,29 +653,29 @@ ; X86-NEXT: pushl %ebx ; X86-NEXT: pushl %edi ; X86-NEXT: pushl %esi +; X86-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-NEXT: movl {{[0-9]+}}(%esp), %edi -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-NEXT: movl {{[0-9]+}}(%esp), %ebx ; X86-NEXT: movl %esi, %eax ; X86-NEXT: shll %cl, %eax ; X86-NEXT: testb $32, %cl -; X86-NEXT: movl $0, %ebx +; X86-NEXT: movl $0, %edi ; X86-NEXT: jne .LBB28_2 ; X86-NEXT: # %bb.1: # %entry -; X86-NEXT: movl %eax, %ebx +; X86-NEXT: movl %eax, %edi ; X86-NEXT: .LBB28_2: # %entry -; X86-NEXT: movl $64, %edx -; X86-NEXT: subl %ecx, %edx -; X86-NEXT: movl %edi, %eax +; X86-NEXT: movb $64, %dl +; X86-NEXT: subb %cl, %dl +; X86-NEXT: movl %ebx, %eax ; X86-NEXT: movl %edx, %ecx ; X86-NEXT: shrl %cl, %eax -; X86-NEXT: shrdl %cl, %edi, %esi +; X86-NEXT: shrdl %cl, %ebx, %esi ; X86-NEXT: testb $32, %dl ; X86-NEXT: jne .LBB28_4 ; X86-NEXT: # %bb.3: # %entry ; X86-NEXT: movl %esi, %eax ; X86-NEXT: .LBB28_4: # %entry -; X86-NEXT: orl %ebx, %eax +; X86-NEXT: orl %edi, %eax ; X86-NEXT: xorl %edx, %edx ; X86-NEXT: popl %esi ; X86-NEXT: popl %edi Index: test/CodeGen/X86/rotate4.ll =================================================================== --- test/CodeGen/X86/rotate4.ll +++ test/CodeGen/X86/rotate4.ll @@ -65,9 +65,9 @@ ; X86-NEXT: .cfi_offset %esi, -16 ; X86-NEXT: .cfi_offset %edi, -12 ; X86-NEXT: .cfi_offset %ebx, -8 +; X86-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi ; X86-NEXT: movl {{[0-9]+}}(%esp), %edi -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X86-NEXT: movl %esi, %eax ; X86-NEXT: shll %cl, %eax ; X86-NEXT: movl %edi, %edx @@ -78,7 +78,7 @@ ; X86-NEXT: movl %eax, %edx ; X86-NEXT: xorl %eax, %eax ; X86-NEXT: .LBB2_2: -; X86-NEXT: negl %ecx +; X86-NEXT: negb %cl ; X86-NEXT: movl %edi, %ebx ; X86-NEXT: shrl %cl, %ebx ; X86-NEXT: shrdl %cl, %edi, %esi @@ -126,9 +126,9 @@ ; X86-NEXT: .cfi_offset %esi, -16 ; X86-NEXT: .cfi_offset %edi, -12 ; X86-NEXT: .cfi_offset %ebx, -8 +; X86-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-NEXT: movl {{[0-9]+}}(%esp), %edi ; X86-NEXT: movl {{[0-9]+}}(%esp), %esi -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X86-NEXT: movl %esi, %edx ; X86-NEXT: shrl %cl, %edx ; X86-NEXT: movl %edi, %eax @@ -139,7 +139,7 @@ ; X86-NEXT: movl %edx, %eax ; X86-NEXT: xorl %edx, %edx ; X86-NEXT: .LBB3_2: -; X86-NEXT: negl %ecx +; X86-NEXT: negb %cl ; X86-NEXT: movl %edi, %ebx ; X86-NEXT: shll %cl, %ebx ; X86-NEXT: shldl %cl, %edi, %esi @@ -242,7 +242,7 @@ ; X86-NEXT: .cfi_offset %edi, -16 ; X86-NEXT: .cfi_offset %ebx, -12 ; X86-NEXT: .cfi_offset %ebp, -8 -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl (%eax), %edx ; X86-NEXT: movl 4(%eax), %ebx @@ -256,7 +256,7 @@ ; X86-NEXT: movl %esi, %edi ; X86-NEXT: xorl %esi, %esi ; X86-NEXT: .LBB6_2: -; X86-NEXT: negl %ecx +; X86-NEXT: negb %cl ; X86-NEXT: movl %ebx, %ebp ; X86-NEXT: shrl %cl, %ebp ; X86-NEXT: shrdl %cl, %ebx, %edx @@ -312,33 +312,33 @@ ; X86-NEXT: .cfi_offset %edi, -16 ; X86-NEXT: .cfi_offset %ebx, -12 ; X86-NEXT: .cfi_offset %ebp, -8 -; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-NEXT: movb {{[0-9]+}}(%esp), %cl ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl (%eax), %ebx -; X86-NEXT: movl 4(%eax), %edx -; X86-NEXT: movl %edx, %esi -; X86-NEXT: shrl %cl, %esi +; X86-NEXT: movl 4(%eax), %esi +; X86-NEXT: movl %esi, %edx +; X86-NEXT: shrl %cl, %edx ; X86-NEXT: movl %ebx, %edi -; X86-NEXT: shrdl %cl, %edx, %edi +; X86-NEXT: shrdl %cl, %esi, %edi ; X86-NEXT: testb $32, %cl ; X86-NEXT: je .LBB7_2 ; X86-NEXT: # %bb.1: -; X86-NEXT: movl %esi, %edi -; X86-NEXT: xorl %esi, %esi +; X86-NEXT: movl %edx, %edi +; X86-NEXT: xorl %edx, %edx ; X86-NEXT: .LBB7_2: -; X86-NEXT: negl %ecx +; X86-NEXT: negb %cl ; X86-NEXT: movl %ebx, %ebp ; X86-NEXT: shll %cl, %ebp -; X86-NEXT: shldl %cl, %ebx, %edx +; X86-NEXT: shldl %cl, %ebx, %esi ; X86-NEXT: testb $32, %cl ; X86-NEXT: je .LBB7_4 ; X86-NEXT: # %bb.3: -; X86-NEXT: movl %ebp, %edx +; X86-NEXT: movl %ebp, %esi ; X86-NEXT: xorl %ebp, %ebp ; X86-NEXT: .LBB7_4: -; X86-NEXT: orl %edx, %esi +; X86-NEXT: orl %esi, %edx ; X86-NEXT: orl %ebp, %edi -; X86-NEXT: movl %esi, 4(%eax) +; X86-NEXT: movl %edx, 4(%eax) ; X86-NEXT: movl %edi, (%eax) ; X86-NEXT: popl %esi ; X86-NEXT: .cfi_def_cfa_offset 16 Index: test/CodeGen/X86/schedule-x86-64-shld.ll =================================================================== --- test/CodeGen/X86/schedule-x86-64-shld.ll +++ test/CodeGen/X86/schedule-x86-64-shld.ll @@ -170,7 +170,7 @@ ; BDVER12-NEXT: movq %rdx, %rcx # sched: [1:0.50] ; BDVER12-NEXT: movq %rsi, %rax # sched: [1:0.50] ; BDVER12-NEXT: shlq %cl, %rdi # sched: [1:0.50] -; BDVER12-NEXT: negl %ecx # sched: [1:0.50] +; BDVER12-NEXT: negb %cl # sched: [1:0.50] ; BDVER12-NEXT: # kill: def $cl killed $cl killed $rcx ; BDVER12-NEXT: shrq %cl, %rax # sched: [1:0.50] ; BDVER12-NEXT: orq %rdi, %rax # sched: [1:0.50] @@ -181,7 +181,7 @@ ; BTVER2-NEXT: movq %rdx, %rcx # sched: [1:0.50] ; BTVER2-NEXT: movq %rsi, %rax # sched: [1:0.50] ; BTVER2-NEXT: shlq %cl, %rdi # sched: [1:0.50] -; BTVER2-NEXT: negl %ecx # sched: [1:0.50] +; BTVER2-NEXT: negb %cl # sched: [1:0.50] ; BTVER2-NEXT: # kill: def $cl killed $cl killed $rcx ; BTVER2-NEXT: shrq %cl, %rax # sched: [1:0.50] ; BTVER2-NEXT: orq %rdi, %rax # sched: [1:0.50] @@ -246,7 +246,7 @@ ; BDVER12-NEXT: movq %rdx, %rcx # sched: [1:0.50] ; BDVER12-NEXT: movq %rsi, %rax # sched: [1:0.50] ; BDVER12-NEXT: shrq %cl, %rdi # sched: [1:0.50] -; BDVER12-NEXT: negl %ecx # sched: [1:0.50] +; BDVER12-NEXT: negb %cl # sched: [1:0.50] ; BDVER12-NEXT: # kill: def $cl killed $cl killed $rcx ; BDVER12-NEXT: shlq %cl, %rax # sched: [1:0.50] ; BDVER12-NEXT: orq %rdi, %rax # sched: [1:0.50] @@ -257,7 +257,7 @@ ; BTVER2-NEXT: movq %rdx, %rcx # sched: [1:0.50] ; BTVER2-NEXT: movq %rsi, %rax # sched: [1:0.50] ; BTVER2-NEXT: shrq %cl, %rdi # sched: [1:0.50] -; BTVER2-NEXT: negl %ecx # sched: [1:0.50] +; BTVER2-NEXT: negb %cl # sched: [1:0.50] ; BTVER2-NEXT: # kill: def $cl killed $cl killed $rcx ; BTVER2-NEXT: shlq %cl, %rax # sched: [1:0.50] ; BTVER2-NEXT: orq %rdi, %rax # sched: [1:0.50] @@ -321,7 +321,7 @@ ; BDVER12-NEXT: movq {{.*}}(%rip), %rax # sched: [5:0.50] ; BDVER12-NEXT: movq %rsi, %rcx # sched: [1:0.50] ; BDVER12-NEXT: shlq %cl, %rax # sched: [1:0.50] -; BDVER12-NEXT: negl %ecx # sched: [1:0.50] +; BDVER12-NEXT: negb %cl # sched: [1:0.50] ; BDVER12-NEXT: # kill: def $cl killed $cl killed $rcx ; BDVER12-NEXT: shrq %cl, %rdi # sched: [1:0.50] ; BDVER12-NEXT: orq %rax, %rdi # sched: [1:0.50] @@ -333,7 +333,7 @@ ; BTVER2-NEXT: movq {{.*}}(%rip), %rax # sched: [5:1.00] ; BTVER2-NEXT: movq %rsi, %rcx # sched: [1:0.50] ; BTVER2-NEXT: shlq %cl, %rax # sched: [1:0.50] -; BTVER2-NEXT: negl %ecx # sched: [1:0.50] +; BTVER2-NEXT: negb %cl # sched: [1:0.50] ; BTVER2-NEXT: # kill: def $cl killed $cl killed $rcx ; BTVER2-NEXT: shrq %cl, %rdi # sched: [1:0.50] ; BTVER2-NEXT: orq %rax, %rdi # sched: [1:0.50] Index: test/CodeGen/X86/scheduler-backtracking.ll =================================================================== --- test/CodeGen/X86/scheduler-backtracking.ll +++ test/CodeGen/X86/scheduler-backtracking.ll @@ -19,78 +19,87 @@ ; ILP-NEXT: pushq %r12 ; ILP-NEXT: pushq %rbx ; ILP-NEXT: movq %rcx, %r9 +; ILP-NEXT: movq %rsi, %rax ; ILP-NEXT: movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill -; ILP-NEXT: xorl %eax, %eax -; ILP-NEXT: addq $1, %rsi +; ILP-NEXT: addq $1, %rax ; ILP-NEXT: adcq $0, %rdx ; ILP-NEXT: adcq $0, %r9 ; ILP-NEXT: adcq $0, %r8 -; ILP-NEXT: leal 1(%rsi,%rsi), %edi -; ILP-NEXT: movl $1, %ebp -; ILP-NEXT: xorl %r14d, %r14d -; ILP-NEXT: movl %edi, %ecx -; ILP-NEXT: shldq %cl, %rbp, %r14 -; ILP-NEXT: movl $1, %r11d -; ILP-NEXT: shlq %cl, %r11 +; ILP-NEXT: movl %eax, %edi +; ILP-NEXT: addb %al, %dil +; ILP-NEXT: orb $1, %dil ; ILP-NEXT: movb $-128, %r10b ; ILP-NEXT: subb %dil, %r10b -; ILP-NEXT: movq %r9, %r13 +; ILP-NEXT: movq %r9, %r12 ; ILP-NEXT: movl %r10d, %ecx -; ILP-NEXT: shlq %cl, %r13 -; ILP-NEXT: movl $1, %r12d -; ILP-NEXT: shrdq %cl, %rax, %r12 -; ILP-NEXT: xorl %r15d, %r15d +; ILP-NEXT: shlq %cl, %r12 +; ILP-NEXT: xorl %ebp, %ebp +; ILP-NEXT: movq %rax, %rsi ; ILP-NEXT: movl %edi, %ecx -; ILP-NEXT: shldq %cl, %r15, %r15 -; ILP-NEXT: movq %rsi, %rbx -; ILP-NEXT: shrdq %cl, %rdx, %rbx +; ILP-NEXT: shrdq %cl, %rdx, %rsi ; ILP-NEXT: shrq %cl, %rdx -; ILP-NEXT: addb $-128, %cl +; ILP-NEXT: movl $1, %ebx +; ILP-NEXT: movl $1, %r13d +; ILP-NEXT: movl %r10d, %ecx +; ILP-NEXT: shrdq %cl, %rbp, %r13 +; ILP-NEXT: testb $64, %r10b +; ILP-NEXT: movl $0, %ecx +; ILP-NEXT: cmovneq %rcx, %r13 +; ILP-NEXT: cmovneq %rcx, %r12 +; ILP-NEXT: xorl %r14d, %r14d +; ILP-NEXT: movl %edi, %ecx +; ILP-NEXT: shldq %cl, %rbx, %r14 +; ILP-NEXT: movl $1, %r15d +; ILP-NEXT: shlq %cl, %r15 +; ILP-NEXT: movl %edi, %r10d +; ILP-NEXT: addb $-128, %r10b +; ILP-NEXT: movl %r10d, %ecx ; ILP-NEXT: shrdq %cl, %r8, %r9 +; ILP-NEXT: xorl %ebp, %ebp +; ILP-NEXT: shldq %cl, %rbx, %rbp +; ILP-NEXT: xorl %r11d, %r11d +; ILP-NEXT: movl %edi, %ecx +; ILP-NEXT: shldq %cl, %r11, %r11 ; ILP-NEXT: testb $64, %dil -; ILP-NEXT: cmovneq %r11, %r14 -; ILP-NEXT: cmoveq %rbx, %rdx -; ILP-NEXT: cmovneq %rax, %r15 -; ILP-NEXT: cmovneq %rax, %r11 -; ILP-NEXT: testb $64, %r10b -; ILP-NEXT: cmovneq %rax, %r12 -; ILP-NEXT: cmovneq %rax, %r13 -; ILP-NEXT: movl $1, %ebx -; ILP-NEXT: shlq %cl, %rbx -; ILP-NEXT: orl %edx, %r13d -; ILP-NEXT: xorl %edx, %edx -; ILP-NEXT: movl $1, %ebp -; ILP-NEXT: shldq %cl, %rbp, %rdx +; ILP-NEXT: cmoveq %rsi, %rdx +; ILP-NEXT: cmovneq %r15, %r14 +; ILP-NEXT: movl $0, %ecx +; ILP-NEXT: cmovneq %rcx, %r15 +; ILP-NEXT: cmovneq %rcx, %r11 +; ILP-NEXT: orl %edx, %r12d +; ILP-NEXT: movl %r10d, %ecx ; ILP-NEXT: shrq %cl, %r8 -; ILP-NEXT: testb $64, %cl +; ILP-NEXT: shlq %cl, %rbx +; ILP-NEXT: xorl %ecx, %ecx +; ILP-NEXT: testb $64, %r10b ; ILP-NEXT: cmoveq %r9, %r8 -; ILP-NEXT: cmovneq %rbx, %rdx -; ILP-NEXT: cmovneq %rax, %rbx +; ILP-NEXT: cmovneq %rbx, %rbp +; ILP-NEXT: cmovneq %rcx, %rbx ; ILP-NEXT: testb %dil, %dil -; ILP-NEXT: cmovsq %rax, %r14 -; ILP-NEXT: cmovsq %rax, %r11 +; ILP-NEXT: cmovsq %rcx, %r14 +; ILP-NEXT: cmovsq %rcx, %r15 ; ILP-NEXT: jns .LBB0_2 ; ILP-NEXT: # %bb.1: -; ILP-NEXT: movl %r8d, %r13d +; ILP-NEXT: movl %r8d, %r12d ; ILP-NEXT: .LBB0_2: ; ILP-NEXT: je .LBB0_4 ; ILP-NEXT: # %bb.3: -; ILP-NEXT: movl %r13d, %esi +; ILP-NEXT: movl %r12d, %eax ; ILP-NEXT: .LBB0_4: -; ILP-NEXT: cmovnsq %r12, %rbx -; ILP-NEXT: cmoveq %rax, %rbx -; ILP-NEXT: cmovnsq %r15, %rdx -; ILP-NEXT: cmoveq %rax, %rdx -; ILP-NEXT: testb $1, %sil -; ILP-NEXT: cmovneq %rax, %rdx +; ILP-NEXT: cmovnsq %r13, %rbx +; ILP-NEXT: cmoveq %rcx, %rbx +; ILP-NEXT: cmovnsq %r11, %rbp +; ILP-NEXT: cmoveq %rcx, %rbp +; ILP-NEXT: testb $1, %al +; ILP-NEXT: cmovneq %rax, %rbp ; ILP-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload -; ILP-NEXT: movq %rdx, 24(%rax) +; ILP-NEXT: movq %rbp, 24(%rax) ; ILP-NEXT: cmovneq %rax, %rbx ; ILP-NEXT: movq %rbx, 16(%rax) ; ILP-NEXT: cmovneq %rax, %r14 ; ILP-NEXT: movq %r14, 8(%rax) -; ILP-NEXT: cmovneq %rax, %r11 -; ILP-NEXT: movq %r11, (%rax) +; ILP-NEXT: cmovneq %rax, %r15 +; ILP-NEXT: movq %r15, (%rax) ; ILP-NEXT: popq %rbx ; ILP-NEXT: popq %r12 ; ILP-NEXT: popq %r13 @@ -114,7 +123,9 @@ ; HYBRID-NEXT: adcq $0, %r9 ; HYBRID-NEXT: adcq $0, %r8 ; HYBRID-NEXT: xorl %r10d, %r10d -; HYBRID-NEXT: leal 1(%rsi,%rsi), %edi +; HYBRID-NEXT: movl %esi, %edi +; HYBRID-NEXT: addb %sil, %dil +; HYBRID-NEXT: orb $1, %dil ; HYBRID-NEXT: xorl %r14d, %r14d ; HYBRID-NEXT: movl %edi, %ecx ; HYBRID-NEXT: shldq %cl, %r14, %r14 @@ -203,7 +214,9 @@ ; BURR-NEXT: adcq $0, %r9 ; BURR-NEXT: adcq $0, %r8 ; BURR-NEXT: xorl %r10d, %r10d -; BURR-NEXT: leal 1(%rsi,%rsi), %edi +; BURR-NEXT: movl %esi, %edi +; BURR-NEXT: addb %sil, %dil +; BURR-NEXT: orb $1, %dil ; BURR-NEXT: xorl %r14d, %r14d ; BURR-NEXT: movl %edi, %ecx ; BURR-NEXT: shldq %cl, %r14, %r14 @@ -291,25 +304,27 @@ ; SRC-NEXT: adcq $0, %rdx ; SRC-NEXT: adcq $0, %r9 ; SRC-NEXT: adcq $0, %r8 -; SRC-NEXT: leal 1(%rsi,%rsi), %r11d +; SRC-NEXT: movl %esi, %edi +; SRC-NEXT: addb %sil, %dil +; SRC-NEXT: orb $1, %dil ; SRC-NEXT: movb $-128, %r10b -; SRC-NEXT: subb %r11b, %r10b +; SRC-NEXT: subb %dil, %r10b ; SRC-NEXT: movq %r9, %r12 ; SRC-NEXT: movl %r10d, %ecx ; SRC-NEXT: shlq %cl, %r12 ; SRC-NEXT: movq %rsi, %rbp -; SRC-NEXT: movl %r11d, %ecx +; SRC-NEXT: movl %edi, %ecx ; SRC-NEXT: shrdq %cl, %rdx, %rbp ; SRC-NEXT: shrq %cl, %rdx ; SRC-NEXT: xorl %r15d, %r15d -; SRC-NEXT: movl $1, %edi +; SRC-NEXT: movl $1, %r11d ; SRC-NEXT: xorl %r14d, %r14d -; SRC-NEXT: shldq %cl, %rdi, %r14 +; SRC-NEXT: shldq %cl, %r11, %r14 ; SRC-NEXT: xorl %r13d, %r13d ; SRC-NEXT: shldq %cl, %r13, %r13 ; SRC-NEXT: movl $1, %ebx ; SRC-NEXT: shlq %cl, %rbx -; SRC-NEXT: testb $64, %r11b +; SRC-NEXT: testb $64, %dil ; SRC-NEXT: cmoveq %rbp, %rdx ; SRC-NEXT: cmovneq %rbx, %r14 ; SRC-NEXT: cmovneq %r15, %rbx @@ -321,18 +336,18 @@ ; SRC-NEXT: cmovneq %r15, %r12 ; SRC-NEXT: cmovneq %r15, %rbp ; SRC-NEXT: orl %edx, %r12d -; SRC-NEXT: movl %r11d, %ecx +; SRC-NEXT: movl %edi, %ecx ; SRC-NEXT: addb $-128, %cl ; SRC-NEXT: shrdq %cl, %r8, %r9 ; SRC-NEXT: shrq %cl, %r8 ; SRC-NEXT: xorl %edx, %edx -; SRC-NEXT: shldq %cl, %rdi, %rdx -; SRC-NEXT: shlq %cl, %rdi +; SRC-NEXT: shldq %cl, %r11, %rdx +; SRC-NEXT: shlq %cl, %r11 ; SRC-NEXT: testb $64, %cl ; SRC-NEXT: cmoveq %r9, %r8 -; SRC-NEXT: cmovneq %rdi, %rdx -; SRC-NEXT: cmovneq %r15, %rdi -; SRC-NEXT: testb %r11b, %r11b +; SRC-NEXT: cmovneq %r11, %rdx +; SRC-NEXT: cmovneq %r15, %r11 +; SRC-NEXT: testb %dil, %dil ; SRC-NEXT: jns .LBB0_2 ; SRC-NEXT: # %bb.1: ; SRC-NEXT: movl %r8d, %r12d @@ -343,17 +358,17 @@ ; SRC-NEXT: .LBB0_4: ; SRC-NEXT: cmovnsq %r13, %rdx ; SRC-NEXT: cmoveq %r15, %rdx -; SRC-NEXT: cmovnsq %rbp, %rdi -; SRC-NEXT: cmoveq %r15, %rdi +; SRC-NEXT: cmovnsq %rbp, %r11 +; SRC-NEXT: cmoveq %r15, %r11 ; SRC-NEXT: cmovsq %r15, %r14 ; SRC-NEXT: cmovsq %r15, %rbx ; SRC-NEXT: testb $1, %sil ; SRC-NEXT: cmovneq %rax, %rbx ; SRC-NEXT: cmovneq %rax, %r14 -; SRC-NEXT: cmovneq %rax, %rdi +; SRC-NEXT: cmovneq %rax, %r11 ; SRC-NEXT: cmovneq %rax, %rdx ; SRC-NEXT: movq %rdx, 24(%rax) -; SRC-NEXT: movq %rdi, 16(%rax) +; SRC-NEXT: movq %r11, 16(%rax) ; SRC-NEXT: movq %r14, 8(%rax) ; SRC-NEXT: movq %rbx, (%rax) ; SRC-NEXT: popq %rbx @@ -376,33 +391,35 @@ ; LIN-NEXT: xorl %r15d, %r15d ; LIN-NEXT: movl $1, %r14d ; LIN-NEXT: addq $1, %rsi -; LIN-NEXT: leal 1(%rsi,%rsi), %ebp +; LIN-NEXT: movl %esi, %r10d +; LIN-NEXT: addb %sil, %r10b +; LIN-NEXT: orb $1, %r10b ; LIN-NEXT: movl $1, %r12d -; LIN-NEXT: movl %ebp, %ecx +; LIN-NEXT: movl %r10d, %ecx ; LIN-NEXT: shlq %cl, %r12 -; LIN-NEXT: testb $64, %bpl +; LIN-NEXT: testb $64, %r10b ; LIN-NEXT: movq %r12, %rbx ; LIN-NEXT: cmovneq %r15, %rbx -; LIN-NEXT: testb %bpl, %bpl +; LIN-NEXT: testb %r10b, %r10b ; LIN-NEXT: cmovsq %r15, %rbx ; LIN-NEXT: adcq $0, %rdx ; LIN-NEXT: adcq $0, %r9 ; LIN-NEXT: adcq $0, %r8 -; LIN-NEXT: movl %ebp, %r10d -; LIN-NEXT: addb $-128, %r10b +; LIN-NEXT: movl %r10d, %ebp +; LIN-NEXT: addb $-128, %bpl ; LIN-NEXT: movq %r9, %rdi -; LIN-NEXT: movl %r10d, %ecx +; LIN-NEXT: movl %ebp, %ecx ; LIN-NEXT: shrdq %cl, %r8, %rdi ; LIN-NEXT: shrq %cl, %r8 -; LIN-NEXT: testb $64, %r10b +; LIN-NEXT: testb $64, %bpl ; LIN-NEXT: cmoveq %rdi, %r8 ; LIN-NEXT: movq %rsi, %rdi -; LIN-NEXT: movl %ebp, %ecx +; LIN-NEXT: movl %r10d, %ecx ; LIN-NEXT: shrdq %cl, %rdx, %rdi ; LIN-NEXT: shrq %cl, %rdx ; LIN-NEXT: cmoveq %rdi, %rdx ; LIN-NEXT: movb $-128, %r11b -; LIN-NEXT: subb %bpl, %r11b +; LIN-NEXT: subb %r10b, %r11b ; LIN-NEXT: movl %r11d, %ecx ; LIN-NEXT: shlq %cl, %r9 ; LIN-NEXT: testb $64, %r11b @@ -420,14 +437,14 @@ ; LIN-NEXT: cmovneq %rax, %rbx ; LIN-NEXT: movq %rbx, (%rax) ; LIN-NEXT: xorl %edx, %edx -; LIN-NEXT: movl %ebp, %ecx +; LIN-NEXT: movl %r10d, %ecx ; LIN-NEXT: shldq %cl, %r14, %rdx ; LIN-NEXT: cmovneq %r12, %rdx ; LIN-NEXT: cmovsq %r15, %rdx ; LIN-NEXT: cmovneq %rax, %rdx ; LIN-NEXT: movq %rdx, 8(%rax) ; LIN-NEXT: movl $1, %edx -; LIN-NEXT: movl %r10d, %ecx +; LIN-NEXT: movl %ebp, %ecx ; LIN-NEXT: shlq %cl, %rdx ; LIN-NEXT: movq %rdx, %rsi ; LIN-NEXT: cmovneq %r15, %rsi @@ -440,11 +457,11 @@ ; LIN-NEXT: cmovneq %rax, %rdi ; LIN-NEXT: movq %rdi, 16(%rax) ; LIN-NEXT: xorl %esi, %esi -; LIN-NEXT: movl %r10d, %ecx +; LIN-NEXT: movl %ebp, %ecx ; LIN-NEXT: shldq %cl, %r14, %rsi ; LIN-NEXT: cmovneq %rdx, %rsi ; LIN-NEXT: xorl %edx, %edx -; LIN-NEXT: movl %ebp, %ecx +; LIN-NEXT: movl %r10d, %ecx ; LIN-NEXT: shldq %cl, %rdx, %rdx ; LIN-NEXT: cmovneq %r15, %rdx ; LIN-NEXT: cmovsq %rsi, %rdx Index: test/CodeGen/X86/test-shrink.ll =================================================================== --- test/CodeGen/X86/test-shrink.ll +++ test/CodeGen/X86/test-shrink.ll @@ -590,7 +590,6 @@ ; CHECK-WIN32-64-LABEL: and16_trunc_8_sign: ; CHECK-WIN32-64: # %bb.0: ; CHECK-WIN32-64-NEXT: subq $40, %rsp -; CHECK-WIN32-64-NEXT: # kill: def $cx killed $cx def $ecx ; CHECK-WIN32-64-NEXT: testb $-128, %cl ; CHECK-WIN32-64-NEXT: jg .LBB13_2 ; CHECK-WIN32-64-NEXT: # %bb.1: # %yes @@ -601,8 +600,7 @@ ; ; CHECK-X86-LABEL: and16_trunc_8_sign: ; CHECK-X86: # %bb.0: -; CHECK-X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax -; CHECK-X86-NEXT: testb $-128, %al +; CHECK-X86-NEXT: testb $-128, {{[0-9]+}}(%esp) ; CHECK-X86-NEXT: jg .LBB13_2 ; CHECK-X86-NEXT: # %bb.1: # %yes ; CHECK-X86-NEXT: calll bar @@ -733,8 +731,8 @@ ; ; CHECK-X86-LABEL: and32_trunc_16_sign: ; CHECK-X86: # %bb.0: -; CHECK-X86-NEXT: movl $32768, %eax # imm = 0x8000 -; CHECK-X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; CHECK-X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax +; CHECK-X86-NEXT: andl $32768, %eax # imm = 0x8000 ; CHECK-X86-NEXT: testw %ax, %ax ; CHECK-X86-NEXT: jg .LBB16_2 ; CHECK-X86-NEXT: # %bb.1: # %yes @@ -778,7 +776,8 @@ ; ; CHECK-X86-LABEL: and32_trunc_16_sign_minsize: ; CHECK-X86: # %bb.0: -; CHECK-X86-NEXT: testw $-32768, {{[0-9]+}}(%esp) # imm = 0x8000 +; CHECK-X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax +; CHECK-X86-NEXT: testw $-32768, %ax # imm = 0x8000 ; CHECK-X86-NEXT: jg .LBB17_2 ; CHECK-X86-NEXT: # %bb.1: # %yes ; CHECK-X86-NEXT: calll bar Index: test/CodeGen/X86/xchg-nofold.ll =================================================================== --- test/CodeGen/X86/xchg-nofold.ll +++ test/CodeGen/X86/xchg-nofold.ll @@ -17,7 +17,7 @@ ; CHECK-NEXT: je .LBB0_3 ; CHECK-NEXT: # %bb.1: ; CHECK-NEXT: movl %edi, %edx -; CHECK-NEXT: andl $7, %edx +; CHECK-NEXT: andb $7, %dl ; CHECK-NEXT: cmpb %cl, %dl ; CHECK-NEXT: jge .LBB0_2 ; CHECK-NEXT: .LBB0_3: