From b3af34d70b4fcc54e72abb9c8b28cfd5c580e599 Mon Sep 17 00:00:00 2001 From: Roman Lebedev Date: Wed, 6 Jun 2018 19:38:16 +0000 Subject: [PATCH] [X86] Emit BZHI when mask is ~(-1 << nbits)) Summary: In D47428, i propose to choose the `~(-(1 << nbits))` as the canonical form of low-bit-mask formation. As it is seen from these tests, there is a reason for that. AArch64 currently better handles `~(-(1 << nbits))`, but not the more traditional `(1 << nbits) - 1` (sic!). The other way around for X86. It would be much better to canonicalize. This patch is completely monkey-typing. I don't really understand how this works :) I have based it on `// x & (-1 >> (32 - y))` pattern. Also, when we only have `BMI`, i wonder if we could use `BEXTR` with `start=0` ? Related links: https://bugs.llvm.org/show_bug.cgi?id=36419 https://bugs.llvm.org/show_bug.cgi?id=37603 https://bugs.llvm.org/show_bug.cgi?id=37610 https://rise4fun.com/Alive/idM Reviewers: craig.topper, spatel, RKSimon, javed.absar Reviewed By: craig.topper Subscribers: kristof.beyls, llvm-commits Differential Revision: https://reviews.llvm.org/D47453 llvm-svn: 334125 --- lib/Target/X86/X86InstrInfo.td | 52 ++++++++--- test/CodeGen/X86/extract-lowbits.ll | 136 ++++++++-------------------- 2 files changed, 76 insertions(+), 112 deletions(-) diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td index 0d2a80f9ace..4815aba5efa 100644 --- a/lib/Target/X86/X86InstrInfo.td +++ b/lib/Target/X86/X86InstrInfo.td @@ -2448,21 +2448,49 @@ let Predicates = [HasBMI2, NoTBM] in { } let Predicates = [HasBMI2] in { - def : Pat<(and GR32:$src, (add (shl 1, GR8:$lz), -1)), - (BZHI32rr GR32:$src, - (INSERT_SUBREG (i32 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>; + multiclass _bmi_bzhi_pattern { + def : Pat; + def : Pat; + } - def : Pat<(and (loadi32 addr:$src), (add (shl 1, GR8:$lz), -1)), - (BZHI32rm addr:$src, - (INSERT_SUBREG (i32 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>; + multiclass bmi_bzhi_patterns { + // x & ((1 << y) - 1) + defm : _bmi_bzhi_pattern<(and RC:$src, (add (shl 1, GR8:$lz), -1)), + (and (x86memop addr:$src), + (add (shl 1, GR8:$lz), -1)), + RC, VT, DstInst, DstMemInst>; - def : Pat<(and GR64:$src, (add (shl 1, GR8:$lz), -1)), - (BZHI64rr GR64:$src, - (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>; + // x & ~(-1 << y) + defm : _bmi_bzhi_pattern<(and RC:$src, (xor (shl -1, GR8:$lz), -1)), + (and (x86memop addr:$src), + (xor (shl -1, GR8:$lz), -1)), + RC, VT, DstInst, DstMemInst>; - def : Pat<(and (loadi64 addr:$src), (add (shl 1, GR8:$lz), -1)), - (BZHI64rm addr:$src, - (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>; + // x & (-1 >> (bitwidth - y)) + defm : _bmi_bzhi_pattern<(and RC:$src, (srl -1, (sub bitwidth, GR8:$lz))), + (and (x86memop addr:$src), + (srl -1, (sub bitwidth, GR8:$lz))), + RC, VT, DstInst, DstMemInst>; + + // x << (bitwidth - y) >> (bitwidth - y) + defm : _bmi_bzhi_pattern<(srl (shl RC:$src, (sub bitwidth, GR8:$lz)), + (sub bitwidth, GR8:$lz)), + (srl (shl (x86memop addr:$src), + (sub bitwidth, GR8:$lz)), + (sub bitwidth, GR8:$lz)), + RC, VT, DstInst, DstMemInst>; + } + + defm : bmi_bzhi_patterns; + defm : bmi_bzhi_patterns; // x & (-1 >> (32 - y)) def : Pat<(and GR32:$src, (srl -1, (i8 (trunc (sub 32, GR32:$lz))))), diff --git a/test/CodeGen/X86/extract-lowbits.ll b/test/CodeGen/X86/extract-lowbits.ll index 825e25fe547..08414d20a49 100644 --- a/test/CodeGen/X86/extract-lowbits.ll +++ b/test/CodeGen/X86/extract-lowbits.ll @@ -537,9 +537,7 @@ define i32 @bzhi32_b0(i32 %val, i32 %numlowbits) nounwind { ; X86-BMI1BMI2-LABEL: bzhi32_b0: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: movl $-1, %ecx -; X86-BMI1BMI2-NEXT: shlxl %eax, %ecx, %eax -; X86-BMI1BMI2-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1BMI2-NEXT: bzhil %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_b0: @@ -553,9 +551,7 @@ define i32 @bzhi32_b0(i32 %val, i32 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_b0: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movl $-1, %eax -; X64-BMI1BMI2-NEXT: shlxl %esi, %eax, %eax -; X64-BMI1BMI2-NEXT: andnl %edi, %eax, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax ; X64-BMI1BMI2-NEXT: retq %notmask = shl i32 -1, %numlowbits %mask = xor i32 %notmask, -1 @@ -576,9 +572,7 @@ define i32 @bzhi32_b1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; X86-BMI1BMI2-LABEL: bzhi32_b1_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: movl $-1, %ecx -; X86-BMI1BMI2-NEXT: shlxl %eax, %ecx, %eax -; X86-BMI1BMI2-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1BMI2-NEXT: bzhil %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_b1_indexzext: @@ -592,9 +586,7 @@ define i32 @bzhi32_b1_indexzext(i32 %val, i8 zeroext %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_b1_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movl $-1, %eax -; X64-BMI1BMI2-NEXT: shlxl %esi, %eax, %eax -; X64-BMI1BMI2-NEXT: andnl %edi, %eax, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax ; X64-BMI1BMI2-NEXT: retq %conv = zext i8 %numlowbits to i32 %notmask = shl i32 -1, %conv @@ -618,9 +610,7 @@ define i32 @bzhi32_b2_load(i32* %w, i32 %numlowbits) nounwind { ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl -; X86-BMI1BMI2-NEXT: movl $-1, %edx -; X86-BMI1BMI2-NEXT: shlxl %ecx, %edx, %ecx -; X86-BMI1BMI2-NEXT: andnl (%eax), %ecx, %eax +; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_b2_load: @@ -634,9 +624,7 @@ define i32 @bzhi32_b2_load(i32* %w, i32 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_b2_load: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movl $-1, %eax -; X64-BMI1BMI2-NEXT: shlxl %esi, %eax, %eax -; X64-BMI1BMI2-NEXT: andnl (%rdi), %eax, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax ; X64-BMI1BMI2-NEXT: retq %val = load i32, i32* %w %notmask = shl i32 -1, %numlowbits @@ -660,9 +648,7 @@ define i32 @bzhi32_b3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl -; X86-BMI1BMI2-NEXT: movl $-1, %edx -; X86-BMI1BMI2-NEXT: shlxl %ecx, %edx, %ecx -; X86-BMI1BMI2-NEXT: andnl (%eax), %ecx, %eax +; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_b3_load_indexzext: @@ -676,9 +662,7 @@ define i32 @bzhi32_b3_load_indexzext(i32* %w, i8 zeroext %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movl $-1, %eax -; X64-BMI1BMI2-NEXT: shlxl %esi, %eax, %eax -; X64-BMI1BMI2-NEXT: andnl (%rdi), %eax, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax ; X64-BMI1BMI2-NEXT: retq %val = load i32, i32* %w %conv = zext i8 %numlowbits to i32 @@ -701,9 +685,7 @@ define i32 @bzhi32_b4_commutative(i32 %val, i32 %numlowbits) nounwind { ; X86-BMI1BMI2-LABEL: bzhi32_b4_commutative: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: movl $-1, %ecx -; X86-BMI1BMI2-NEXT: shlxl %eax, %ecx, %eax -; X86-BMI1BMI2-NEXT: andnl {{[0-9]+}}(%esp), %eax, %eax +; X86-BMI1BMI2-NEXT: bzhil %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_b4_commutative: @@ -717,9 +699,7 @@ define i32 @bzhi32_b4_commutative(i32 %val, i32 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_b4_commutative: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movl $-1, %eax -; X64-BMI1BMI2-NEXT: shlxl %esi, %eax, %eax -; X64-BMI1BMI2-NEXT: andnl %edi, %eax, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax ; X64-BMI1BMI2-NEXT: retq %notmask = shl i32 -1, %numlowbits %mask = xor i32 %notmask, -1 @@ -778,9 +758,7 @@ define i64 @bzhi64_b0(i64 %val, i64 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_b0: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movq $-1, %rax -; X64-BMI1BMI2-NEXT: shlxq %rsi, %rax, %rax -; X64-BMI1BMI2-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax ; X64-BMI1BMI2-NEXT: retq %notmask = shl i64 -1, %numlowbits %mask = xor i64 %notmask, -1 @@ -838,9 +816,7 @@ define i64 @bzhi64_b1_indexzext(i64 %val, i8 zeroext %numlowbits) nounwind { ; X64-BMI1BMI2-LABEL: bzhi64_b1_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi -; X64-BMI1BMI2-NEXT: movq $-1, %rax -; X64-BMI1BMI2-NEXT: shlxq %rsi, %rax, %rax -; X64-BMI1BMI2-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax ; X64-BMI1BMI2-NEXT: retq %conv = zext i8 %numlowbits to i64 %notmask = shl i64 -1, %conv @@ -902,9 +878,7 @@ define i64 @bzhi64_b2_load(i64* %w, i64 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_b2_load: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movq $-1, %rax -; X64-BMI1BMI2-NEXT: shlxq %rsi, %rax, %rax -; X64-BMI1BMI2-NEXT: andnq (%rdi), %rax, %rax +; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax ; X64-BMI1BMI2-NEXT: retq %val = load i64, i64* %w %notmask = shl i64 -1, %numlowbits @@ -967,9 +941,7 @@ define i64 @bzhi64_b3_load_indexzext(i64* %w, i8 zeroext %numlowbits) nounwind { ; X64-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi -; X64-BMI1BMI2-NEXT: movq $-1, %rax -; X64-BMI1BMI2-NEXT: shlxq %rsi, %rax, %rax -; X64-BMI1BMI2-NEXT: andnq (%rdi), %rax, %rax +; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax ; X64-BMI1BMI2-NEXT: retq %val = load i64, i64* %w %conv = zext i8 %numlowbits to i64 @@ -1028,9 +1000,7 @@ define i64 @bzhi64_b4_commutative(i64 %val, i64 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_b4_commutative: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movq $-1, %rax -; X64-BMI1BMI2-NEXT: shlxq %rsi, %rax, %rax -; X64-BMI1BMI2-NEXT: andnq %rdi, %rax, %rax +; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax ; X64-BMI1BMI2-NEXT: retq %notmask = shl i64 -1, %numlowbits %mask = xor i64 %notmask, -1 @@ -1091,11 +1061,8 @@ define i32 @bzhi32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; ; X86-BMI1BMI2-LABEL: bzhi32_c1_indexzext: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movb $32, %al -; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: movl $-1, %ecx -; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax -; X86-BMI1BMI2-NEXT: andl {{[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 ; ; X64-NOBMI-LABEL: bzhi32_c1_indexzext: @@ -1109,11 +1076,7 @@ define i32 @bzhi32_c1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_c1_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $32, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: movl $-1, %ecx -; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax -; X64-BMI1BMI2-NEXT: andl %edi, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax ; X64-BMI1BMI2-NEXT: retq %numhighbits = sub i8 32, %numlowbits %sh_prom = zext i8 %numhighbits to i32 @@ -1175,12 +1138,9 @@ define i32 @bzhi32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; ; X86-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-BMI1BMI2-NEXT: movb $32, %al -; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: movl $-1, %edx -; X86-BMI1BMI2-NEXT: shrxl %eax, %edx, %eax -; X86-BMI1BMI2-NEXT: andl (%ecx), %eax +; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_c3_load_indexzext: @@ -1194,11 +1154,7 @@ define i32 @bzhi32_c3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $32, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: movl $-1, %ecx -; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax -; X64-BMI1BMI2-NEXT: andl (%rdi), %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax ; X64-BMI1BMI2-NEXT: retq %val = load i32, i32* %w %numhighbits = sub i8 32, %numlowbits @@ -1350,11 +1306,8 @@ define i64 @bzhi64_c1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_c1_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $64, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: movq $-1, %rcx -; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rax -; X64-BMI1BMI2-NEXT: andq %rdi, %rax +; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi +; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax ; X64-BMI1BMI2-NEXT: retq %numhighbits = sub i8 64, %numlowbits %sh_prom = zext i8 %numhighbits to i64 @@ -1479,11 +1432,8 @@ define i64 @bzhi64_c3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $64, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: movq $-1, %rcx -; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rax -; X64-BMI1BMI2-NEXT: andq (%rdi), %rax +; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi +; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax ; X64-BMI1BMI2-NEXT: retq %val = load i64, i64* %w %numhighbits = sub i8 64, %numlowbits @@ -1602,10 +1552,8 @@ define i32 @bzhi32_d1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; ; X86-BMI1BMI2-LABEL: bzhi32_d1_indexzext: ; X86-BMI1BMI2: # %bb.0: -; X86-BMI1BMI2-NEXT: movb $32, %al -; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %al -; X86-BMI1BMI2-NEXT: shlxl %eax, {{[0-9]+}}(%esp), %ecx -; X86-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al +; X86-BMI1BMI2-NEXT: bzhil %eax, {{[0-9]+}}(%esp), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_d1_indexzext: @@ -1619,10 +1567,7 @@ define i32 @bzhi32_d1_indexzext(i32 %val, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_d1_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $32, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: shlxl %eax, %edi, %ecx -; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, %edi, %eax ; X64-BMI1BMI2-NEXT: retq %numhighbits = sub i8 32, %numlowbits %sh_prom = zext i8 %numhighbits to i32 @@ -1685,10 +1630,8 @@ define i32 @bzhi32_d3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; X86-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext: ; X86-BMI1BMI2: # %bb.0: ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-BMI1BMI2-NEXT: movb $32, %cl -; X86-BMI1BMI2-NEXT: subb {{[0-9]+}}(%esp), %cl -; X86-BMI1BMI2-NEXT: shlxl %ecx, (%eax), %eax -; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %eax +; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl +; X86-BMI1BMI2-NEXT: bzhil %ecx, (%eax), %eax ; X86-BMI1BMI2-NEXT: retl ; ; X64-NOBMI-LABEL: bzhi32_d3_load_indexzext: @@ -1702,10 +1645,7 @@ define i32 @bzhi32_d3_load_indexzext(i32* %w, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $32, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: shlxl %eax, (%rdi), %ecx -; X64-BMI1BMI2-NEXT: shrxl %eax, %ecx, %eax +; X64-BMI1BMI2-NEXT: bzhil %esi, (%rdi), %eax ; X64-BMI1BMI2-NEXT: retq %val = load i32, i32* %w %numhighbits = sub i8 32, %numlowbits @@ -1892,10 +1832,8 @@ define i64 @bzhi64_d1_indexzext(i64 %val, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_d1_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $64, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: shlxq %rax, %rdi, %rcx -; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rax +; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi +; X64-BMI1BMI2-NEXT: bzhiq %rsi, %rdi, %rax ; X64-BMI1BMI2-NEXT: retq %numhighbits = sub i8 64, %numlowbits %sh_prom = zext i8 %numhighbits to i64 @@ -2084,10 +2022,8 @@ define i64 @bzhi64_d3_load_indexzext(i64* %w, i8 %numlowbits) nounwind { ; ; X64-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext: ; X64-BMI1BMI2: # %bb.0: -; X64-BMI1BMI2-NEXT: movb $64, %al -; X64-BMI1BMI2-NEXT: subb %sil, %al -; X64-BMI1BMI2-NEXT: shlxq %rax, (%rdi), %rcx -; X64-BMI1BMI2-NEXT: shrxq %rax, %rcx, %rax +; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi +; X64-BMI1BMI2-NEXT: bzhiq %rsi, (%rdi), %rax ; X64-BMI1BMI2-NEXT: retq %val = load i64, i64* %w %numhighbits = sub i8 64, %numlowbits