| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI,X86-NOBMI |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1NOTBM,X86-BMI1NOTBM |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1TBM,X86-BMI1TBM |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1TBM,X86-BMI1TBM,BMI1TBMBMI2,X86-BMI1TBMBMI2 |
| ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1NOTBMBMI2,X86-BMI1NOTBMBMI2 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI,X64-NOBMI |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1NOTBM,X64-BMI1NOTBM |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1TBM,X64-BMI1TBM |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1TBM,X64-BMI1TBM,BMI1TBMBMI2,X64-BMI1TBMBMI2 |
| ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2,+fast-bextr < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1NOTBMBMI2,X64-BMI1NOTBMBMI2 |
| |
| ; *Please* keep in sync with test/CodeGen/AArch64/extract-bits.ll |
| |
| ; 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 |
| |
| ; Patterns: |
| ; a) (x >> start) & (1 << nbits) - 1 |
| ; b) (x >> start) & ~(-1 << nbits) |
| ; c) (x >> start) & (-1 >> (32 - y)) |
| ; d) (x >> start) << (32 - y) >> (32 - y) |
| ; are equivalent. |
| |
| declare void @use32(i32) |
| declare void @use64(i64) |
| |
| ; ---------------------------------------------------------------------------- ; |
| ; Pattern a. 32-bit |
| ; ---------------------------------------------------------------------------- ; |
| |
| define i32 @bextr32_a0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_a0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a0_arithmetic(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a0_arithmetic: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: sarl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a0_arithmetic: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1NOTBM-NEXT: sarl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shll $8, %eax |
| ; X86-BMI1NOTBM-NEXT: bextrl %eax, %edx, %eax |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr32_a0_arithmetic: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: sarxl %ecx, {{[0-9]+}}(%esp), %ecx |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a0_arithmetic: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: sarl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a0_arithmetic: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movl %esi, %ecx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-BMI1NOTBM-NEXT: sarl %cl, %edi |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: bextrl %edx, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a0_arithmetic: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: sarxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = ashr i32 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a1_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a1_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_a1_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a1_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a1_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a1_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %skip = zext i8 %numskipbits to i32 |
| %shifted = lshr i32 %val, %skip |
| %conv = zext i8 %numlowbits to i32 |
| %onebit = shl i32 1, %conv |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a2_load: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a2_load: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: shll $8, %ecx |
| ; 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_a2_load: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a2_load: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movl (%rdi), %esi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %esi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %esi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a2_load: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, (%rdi), %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a2_load: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, (%rdi), %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i32, i32* %w |
| %shifted = lshr i32 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a3_load_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: shll $8, %ecx |
| ; 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_a3_load_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a3_load_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movl (%rdi), %esi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %esi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %esi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, (%rdi), %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a3_load_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, (%rdi), %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i32, i32* %w |
| %skip = zext i8 %numskipbits to i32 |
| %shifted = lshr i32 %val, %skip |
| %conv = zext i8 %numlowbits to i32 |
| %onebit = shl i32 1, %conv |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a4_commutative: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a4_commutative: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_a4_commutative: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a4_commutative: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a4_commutative: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a4_commutative: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %shifted, %mask ; swapped order |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_a5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_a5_skipextrauses: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: pushl %eax |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %ecx |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: movl $1, %esi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: decl %esi |
| ; X86-NOBMI-NEXT: andl %edi, %esi |
| ; X86-NOBMI-NEXT: movl %eax, (%esp) |
| ; X86-NOBMI-NEXT: calll use32 |
| ; X86-NOBMI-NEXT: movl %esi, %eax |
| ; X86-NOBMI-NEXT: addl $4, %esp |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_a5_skipextrauses: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: subl $8, %esp |
| ; 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 |
| ; X86-BMI1NOTBM-NEXT: orl %ecx, %edx |
| ; X86-BMI1NOTBM-NEXT: bextrl %edx, {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl %eax, (%esp) |
| ; X86-BMI1NOTBM-NEXT: calll use32 |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: addl $8, %esp |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr32_a5_skipextrauses: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: subl $8, %esp |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: movl %ecx, (%esp) |
| ; X86-BMI1BMI2-NEXT: calll use32 |
| ; X86-BMI1BMI2-NEXT: movl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: addl $8, %esp |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_a5_skipextrauses: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: pushq %rbx |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $1, %ebx |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %ebx |
| ; X64-NOBMI-NEXT: decl %ebx |
| ; X64-NOBMI-NEXT: andl %edi, %ebx |
| ; X64-NOBMI-NEXT: movl %esi, %edi |
| ; X64-NOBMI-NEXT: callq use32 |
| ; X64-NOBMI-NEXT: movl %ebx, %eax |
| ; X64-NOBMI-NEXT: popq %rbx |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_a5_skipextrauses: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: pushq %rbx |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %ebx |
| ; X64-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X64-BMI1NOTBM-NEXT: callq use32 |
| ; X64-BMI1NOTBM-NEXT: movl %ebx, %eax |
| ; X64-BMI1NOTBM-NEXT: popq %rbx |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_a5_skipextrauses: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: pushq %rbx |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %ebx |
| ; X64-BMI1BMI2-NEXT: movl %esi, %edi |
| ; X64-BMI1BMI2-NEXT: callq use32 |
| ; X64-BMI1BMI2-NEXT: movl %ebx, %eax |
| ; X64-BMI1BMI2-NEXT: popq %rbx |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %shifted |
| call void @use32(i32 %numskipbits) |
| ret i32 %masked |
| } |
| |
| ; 64-bit |
| |
| define i64 @bextr64_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB7_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB7_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %eax, %edx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB7_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB7_4: |
| ; X86-NOBMI-NEXT: addl $-1, %eax |
| ; X86-NOBMI-NEXT: adcl $-1, %edx |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edi |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB7_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: .LBB7_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB7_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB7_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %eax |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edx |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB7_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edi, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB7_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB7_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edx |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB7_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %eax |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %edx |
| ; X86-BMI1BMI2-NEXT: andl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: andl %edi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a0_arithmetic(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a0_arithmetic: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %esi |
| ; X86-NOBMI-NEXT: sarl %cl, %esi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %edi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB8_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: sarl $31, %eax |
| ; X86-NOBMI-NEXT: movl %esi, %edi |
| ; X86-NOBMI-NEXT: movl %eax, %esi |
| ; X86-NOBMI-NEXT: .LBB8_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %eax, %edx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB8_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB8_4: |
| ; X86-NOBMI-NEXT: addl $-1, %eax |
| ; X86-NOBMI-NEXT: adcl $-1, %edx |
| ; X86-NOBMI-NEXT: andl %edi, %eax |
| ; X86-NOBMI-NEXT: andl %esi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a0_arithmetic: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: sarl %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB8_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: sarl $31, %eax |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: .LBB8_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB8_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB8_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %eax |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edx |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a0_arithmetic: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1BMI2-NEXT: sarxl %ecx, %eax, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB8_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: sarl $31, %eax |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB8_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB8_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edx |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB8_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %eax |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %edx |
| ; X86-BMI1BMI2-NEXT: andl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: andl %edi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a0_arithmetic: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: sarq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a0_arithmetic: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-BMI1NOTBM-NEXT: sarq %cl, %rdi |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: bextrq %rdx, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a0_arithmetic: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: sarxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = ashr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a1_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB9_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB9_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %eax, %edx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB9_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB9_4: |
| ; X86-NOBMI-NEXT: addl $-1, %eax |
| ; X86-NOBMI-NEXT: adcl $-1, %edx |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a1_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edi |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB9_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: .LBB9_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB9_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB9_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %eax |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edx |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a1_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB9_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edi, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB9_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB9_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edx |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB9_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %eax |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %edx |
| ; X86-BMI1BMI2-NEXT: andl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: andl %edi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a1_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a1_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a1_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %skip = zext i8 %numskipbits to i64 |
| %shifted = lshr i64 %val, %skip |
| %conv = zext i8 %numlowbits to i64 |
| %onebit = shl i64 1, %conv |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a2_load: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: movl 4(%eax), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB10_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB10_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %eax, %edx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB10_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB10_4: |
| ; X86-NOBMI-NEXT: addl $-1, %eax |
| ; X86-NOBMI-NEXT: adcl $-1, %edx |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a2_load: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl (%eax), %esi |
| ; X86-BMI1NOTBM-NEXT: movl 4(%eax), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edi |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB10_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: .LBB10_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB10_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB10_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %eax |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edx |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a2_load: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: movl (%eax), %esi |
| ; X86-BMI1BMI2-NEXT: movl 4(%eax), %eax |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %edi |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB10_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edi, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB10_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB10_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edx |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB10_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %eax |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %edx |
| ; X86-BMI1BMI2-NEXT: andl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: andl %edi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a2_load: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: movq (%rdi), %rsi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rsi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rsi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a2_load: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, (%rdi), %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a2_load: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, (%rdi), %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i64, i64* %w |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a3_load_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: movl 4(%eax), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB11_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB11_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %eax, %edx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB11_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB11_4: |
| ; X86-NOBMI-NEXT: addl $-1, %eax |
| ; X86-NOBMI-NEXT: adcl $-1, %edx |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl (%eax), %esi |
| ; X86-BMI1NOTBM-NEXT: movl 4(%eax), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edi |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB11_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: .LBB11_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB11_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %edx |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB11_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %eax |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edx |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a3_load_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: movl (%eax), %esi |
| ; X86-BMI1BMI2-NEXT: movl 4(%eax), %eax |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %edi |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %esi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB11_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edi, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB11_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB11_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %eax, %edx |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB11_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %eax |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %edx |
| ; X86-BMI1BMI2-NEXT: andl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: andl %edi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a3_load_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movq (%rdi), %rsi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rsi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rsi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, (%rdi), %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a3_load_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, (%rdi), %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i64, i64* %w |
| %skip = zext i8 %numskipbits to i64 |
| %shifted = lshr i64 %val, %skip |
| %conv = zext i8 %numlowbits to i64 |
| %onebit = shl i64 1, %conv |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a4_commutative: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl %esi, %edx |
| ; X86-NOBMI-NEXT: shrl %cl, %edx |
| ; X86-NOBMI-NEXT: shrdl %cl, %esi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB12_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edx, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: .LBB12_2: |
| ; X86-NOBMI-NEXT: movl $1, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shldl %cl, %esi, %edi |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB12_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %esi, %edi |
| ; X86-NOBMI-NEXT: xorl %esi, %esi |
| ; X86-NOBMI-NEXT: .LBB12_4: |
| ; X86-NOBMI-NEXT: addl $-1, %esi |
| ; X86-NOBMI-NEXT: adcl $-1, %edi |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a4_commutative: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB12_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %eax |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB12_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: movb %ch, %cl |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %ch |
| ; X86-BMI1NOTBM-NEXT: je .LBB12_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %esi, %esi |
| ; X86-BMI1NOTBM-NEXT: .LBB12_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %esi |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: andl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a4_commutative: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edx, %eax |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edx, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB12_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %eax |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB12_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %esi, %esi |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %esi |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %edi, %ecx |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: je .LBB12_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ecx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %ecx, %ecx |
| ; X86-BMI1BMI2-NEXT: .LBB12_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %ecx |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %esi |
| ; X86-BMI1BMI2-NEXT: andl %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: andl %esi, %edx |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a4_commutative: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a4_commutative: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a4_commutative: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %shifted, %mask ; swapped order |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_a5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_a5_skipextrauses: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %ebp |
| ; X86-NOBMI-NEXT: pushl %ebx |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: subl $12, %esp |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %ebx |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %esi, %ebp |
| ; X86-NOBMI-NEXT: movl %eax, %ecx |
| ; X86-NOBMI-NEXT: shrl %cl, %ebp |
| ; X86-NOBMI-NEXT: shrdl %cl, %esi, %ebx |
| ; X86-NOBMI-NEXT: testb $32, %al |
| ; X86-NOBMI-NEXT: je .LBB13_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %ebp, %ebx |
| ; X86-NOBMI-NEXT: xorl %ebp, %ebp |
| ; X86-NOBMI-NEXT: .LBB13_2: |
| ; X86-NOBMI-NEXT: movl $1, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shldl %cl, %esi, %edi |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: testb $32, %dl |
| ; X86-NOBMI-NEXT: je .LBB13_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %esi, %edi |
| ; X86-NOBMI-NEXT: xorl %esi, %esi |
| ; X86-NOBMI-NEXT: .LBB13_4: |
| ; X86-NOBMI-NEXT: addl $-1, %esi |
| ; X86-NOBMI-NEXT: adcl $-1, %edi |
| ; X86-NOBMI-NEXT: andl %ebx, %esi |
| ; X86-NOBMI-NEXT: andl %ebp, %edi |
| ; X86-NOBMI-NEXT: subl $8, %esp |
| ; X86-NOBMI-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-NOBMI-NEXT: pushl %eax |
| ; X86-NOBMI-NEXT: calll use64 |
| ; X86-NOBMI-NEXT: addl $16, %esp |
| ; X86-NOBMI-NEXT: movl %esi, %eax |
| ; X86-NOBMI-NEXT: movl %edi, %edx |
| ; X86-NOBMI-NEXT: addl $12, %esp |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: popl %ebx |
| ; X86-NOBMI-NEXT: popl %ebp |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebp |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: subl $12, %esp |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %ebx |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %ebp |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %ebp |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %esi, %ebx |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB13_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %ebp, %ebx |
| ; X86-BMI1NOTBM-NEXT: xorl %ebp, %ebp |
| ; X86-BMI1NOTBM-NEXT: .LBB13_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %ecx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %dl |
| ; X86-BMI1NOTBM-NEXT: je .LBB13_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %esi, %esi |
| ; X86-BMI1NOTBM-NEXT: .LBB13_4: |
| ; X86-BMI1NOTBM-NEXT: addl $-1, %esi |
| ; X86-BMI1NOTBM-NEXT: adcl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: andl %ebx, %esi |
| ; X86-BMI1NOTBM-NEXT: andl %ebp, %edi |
| ; X86-BMI1NOTBM-NEXT: subl $8, %esp |
| ; X86-BMI1NOTBM-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-BMI1NOTBM-NEXT: pushl %eax |
| ; X86-BMI1NOTBM-NEXT: calll use64 |
| ; X86-BMI1NOTBM-NEXT: addl $16, %esp |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: addl $12, %esp |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: popl %ebp |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_a5_skipextrauses: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebp |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: subl $12, %esp |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %ebx |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %ebx |
| ; X86-BMI1BMI2-NEXT: shrxl %eax, %esi, %ebp |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB13_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %ebp, %ebx |
| ; X86-BMI1BMI2-NEXT: xorl %ebp, %ebp |
| ; X86-BMI1BMI2-NEXT: .LBB13_2: |
| ; X86-BMI1BMI2-NEXT: movl $1, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %esi, %esi |
| ; X86-BMI1BMI2-NEXT: movl %edx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %esi |
| ; X86-BMI1BMI2-NEXT: shlxl %edx, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %dl |
| ; X86-BMI1BMI2-NEXT: je .LBB13_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %edi, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edi, %edi |
| ; X86-BMI1BMI2-NEXT: .LBB13_4: |
| ; X86-BMI1BMI2-NEXT: addl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: adcl $-1, %esi |
| ; X86-BMI1BMI2-NEXT: andl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: andl %ebp, %esi |
| ; X86-BMI1BMI2-NEXT: subl $8, %esp |
| ; X86-BMI1BMI2-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-BMI1BMI2-NEXT: pushl %eax |
| ; X86-BMI1BMI2-NEXT: calll use64 |
| ; X86-BMI1BMI2-NEXT: addl $16, %esp |
| ; X86-BMI1BMI2-NEXT: movl %edi, %eax |
| ; X86-BMI1BMI2-NEXT: movl %esi, %edx |
| ; X86-BMI1BMI2-NEXT: addl $12, %esp |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: popl %ebp |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_a5_skipextrauses: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: pushq %rbx |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %ebx |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rbx |
| ; X64-NOBMI-NEXT: decq %rbx |
| ; X64-NOBMI-NEXT: andq %rdi, %rbx |
| ; X64-NOBMI-NEXT: movq %rsi, %rdi |
| ; X64-NOBMI-NEXT: callq use64 |
| ; X64-NOBMI-NEXT: movq %rbx, %rax |
| ; X64-NOBMI-NEXT: popq %rbx |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: pushq %rbx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rbx |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rdi |
| ; X64-BMI1NOTBM-NEXT: callq use64 |
| ; X64-BMI1NOTBM-NEXT: movq %rbx, %rax |
| ; X64-BMI1NOTBM-NEXT: popq %rbx |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_a5_skipextrauses: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: pushq %rbx |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rbx |
| ; X64-BMI1BMI2-NEXT: movq %rsi, %rdi |
| ; X64-BMI1BMI2-NEXT: callq use64 |
| ; X64-BMI1BMI2-NEXT: movq %rbx, %rax |
| ; X64-BMI1BMI2-NEXT: popq %rbx |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| call void @use64(i64 %numskipbits) |
| ret i64 %masked |
| } |
| |
| ; 64-bit, but with 32-bit output |
| |
| ; Everything done in 64-bit, truncation happens last. |
| define i32 @bextr64_32_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_32_a0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: shrdl %cl, %edi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: jne .LBB14_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %eax, %esi |
| ; X86-NOBMI-NEXT: .LBB14_2: |
| ; X86-NOBMI-NEXT: movl $1, %edi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %edi |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: testb $32, %dl |
| ; X86-NOBMI-NEXT: jne .LBB14_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %edi, %eax |
| ; X86-NOBMI-NEXT: .LBB14_4: |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_32_a0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: jne .LBB14_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %esi |
| ; X86-BMI1NOTBM-NEXT: .LBB14_2: |
| ; X86-BMI1NOTBM-NEXT: movl $1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %eax, %eax |
| ; X86-BMI1NOTBM-NEXT: testb $32, %dl |
| ; X86-BMI1NOTBM-NEXT: jne .LBB14_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %eax |
| ; X86-BMI1NOTBM-NEXT: .LBB14_4: |
| ; X86-BMI1NOTBM-NEXT: decl %eax |
| ; X86-BMI1NOTBM-NEXT: andl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_32_a0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %bl |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB14_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %eax, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB14_2: |
| ; X86-BMI1BMI2-NEXT: xorl %eax, %eax |
| ; X86-BMI1BMI2-NEXT: testb $32, %bl |
| ; X86-BMI1BMI2-NEXT: jne .LBB14_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl $1, %eax |
| ; X86-BMI1BMI2-NEXT: shlxl %ebx, %eax, %eax |
| ; X86-BMI1BMI2-NEXT: .LBB14_4: |
| ; X86-BMI1BMI2-NEXT: decl %eax |
| ; X86-BMI1BMI2-NEXT: andl %edx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_32_a0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: # kill: def $eax killed $eax killed $rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_32_a0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-BMI1NOTBM-NEXT: shrq %cl, %rdi |
| ; X64-BMI1NOTBM-NEXT: movl $1, %eax |
| ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx |
| ; X64-BMI1NOTBM-NEXT: shlq %cl, %rax |
| ; X64-BMI1NOTBM-NEXT: decl %eax |
| ; X64-BMI1NOTBM-NEXT: andl %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: # kill: def $eax killed $eax killed $rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_32_a0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: movl $1, %ecx |
| ; X64-BMI1BMI2-NEXT: shlxq %rdx, %rcx, %rcx |
| ; X64-BMI1BMI2-NEXT: decl %ecx |
| ; X64-BMI1BMI2-NEXT: andl %ecx, %eax |
| ; X64-BMI1BMI2-NEXT: # kill: def $eax killed $eax killed $rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i64 1, %numlowbits |
| %mask = add nsw i64 %onebit, -1 |
| %masked = and i64 %mask, %shifted |
| %res = trunc i64 %masked to i32 |
| ret i32 %res |
| } |
| |
| ; Shifting happens in 64-bit, then truncation. Masking is 32-bit. |
| define i32 @bextr64_32_a1(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_32_a1: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: shrdl %cl, %edi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: jne .LBB15_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %eax, %esi |
| ; X86-NOBMI-NEXT: .LBB15_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_32_a1: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: jne .LBB15_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB15_2: |
| ; X86-BMI1NOTBM-NEXT: shll $8, %eax |
| ; X86-BMI1NOTBM-NEXT: bextrl %eax, %edx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_32_a1: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB15_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB15_2: |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %edx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_32_a1: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_32_a1: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-BMI1NOTBM-NEXT: shrq %cl, %rdi |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: bextrl %edx, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_32_a1: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %truncshifted = trunc i64 %shifted to i32 |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %masked = and i32 %mask, %truncshifted |
| ret i32 %masked |
| } |
| |
| ; Shifting happens in 64-bit. Mask is 32-bit, but extended to 64-bit. |
| ; Masking is 64-bit. Then truncation. |
| define i32 @bextr64_32_a2(i64 %val, i64 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_32_a2: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: shrdl %cl, %edi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: jne .LBB16_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %eax, %esi |
| ; X86-NOBMI-NEXT: .LBB16_2: |
| ; X86-NOBMI-NEXT: movl $1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: decl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_32_a2: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: jne .LBB16_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB16_2: |
| ; X86-BMI1NOTBM-NEXT: shll $8, %eax |
| ; X86-BMI1NOTBM-NEXT: bextrl %eax, %edx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_32_a2: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB16_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB16_2: |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %edx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_32_a2: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movl $1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: decl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_32_a2: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-BMI1NOTBM-NEXT: shrq %cl, %rdi |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: bextrl %edx, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_32_a2: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %onebit = shl i32 1, %numlowbits |
| %mask = add nsw i32 %onebit, -1 |
| %zextmask = zext i32 %mask to i64 |
| %masked = and i64 %zextmask, %shifted |
| %truncmasked = trunc i64 %masked to i32 |
| ret i32 %truncmasked |
| } |
| |
| ; ---------------------------------------------------------------------------- ; |
| ; Pattern b. 32-bit |
| ; ---------------------------------------------------------------------------- ; |
| |
| define i32 @bextr32_b0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_b0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $-1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %notmask = shl i32 -1, %numlowbits |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_b1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b1_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b1_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_b1_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b1_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $-1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b1_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b1_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %skip = zext i8 %numskipbits to i32 |
| %shifted = lshr i32 %val, %skip |
| %conv = zext i8 %numlowbits to i32 |
| %notmask = shl i32 -1, %conv |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_b2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b2_load: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b2_load: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: shll $8, %ecx |
| ; 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_b2_load: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b2_load: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movl (%rdi), %esi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %esi |
| ; X64-NOBMI-NEXT: movl $-1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %esi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b2_load: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, (%rdi), %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b2_load: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, (%rdi), %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i32, i32* %w |
| %shifted = lshr i32 %val, %numskipbits |
| %notmask = shl i32 -1, %numlowbits |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_b3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b3_load_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: shll $8, %ecx |
| ; 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_b3_load_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: bzhil %eax, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b3_load_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movl (%rdi), %esi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %esi |
| ; X64-NOBMI-NEXT: movl $-1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %esi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, (%rdi), %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b3_load_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, (%rdi), %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i32, i32* %w |
| %skip = zext i8 %numskipbits to i32 |
| %shifted = lshr i32 %val, %skip |
| %conv = zext i8 %numlowbits to i32 |
| %notmask = shl i32 -1, %conv |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %mask, %shifted |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_b4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b4_commutative: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b4_commutative: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: shll $8, %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_b4_commutative: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b4_commutative: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $-1, %eax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %eax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b4_commutative: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b4_commutative: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %notmask = shl i32 -1, %numlowbits |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %shifted, %mask ; swapped order |
| ret i32 %masked |
| } |
| |
| define i32 @bextr32_b5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr32_b5_skipextrauses: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: pushl %eax |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %ecx |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: movl $-1, %esi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: notl %esi |
| ; X86-NOBMI-NEXT: andl %edi, %esi |
| ; X86-NOBMI-NEXT: movl %eax, (%esp) |
| ; X86-NOBMI-NEXT: calll use32 |
| ; X86-NOBMI-NEXT: movl %esi, %eax |
| ; X86-NOBMI-NEXT: addl $4, %esp |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: subl $8, %esp |
| ; 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 |
| ; X86-BMI1NOTBM-NEXT: orl %ecx, %edx |
| ; X86-BMI1NOTBM-NEXT: bextrl %edx, {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl %eax, (%esp) |
| ; X86-BMI1NOTBM-NEXT: calll use32 |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %eax |
| ; X86-BMI1NOTBM-NEXT: addl $8, %esp |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr32_b5_skipextrauses: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: subl $8, %esp |
| ; 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 |
| ; X86-BMI1BMI2-NEXT: movl %ecx, (%esp) |
| ; X86-BMI1BMI2-NEXT: calll use32 |
| ; X86-BMI1BMI2-NEXT: movl %esi, %eax |
| ; X86-BMI1BMI2-NEXT: addl $8, %esp |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr32_b5_skipextrauses: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: pushq %rbx |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: shrl %cl, %edi |
| ; X64-NOBMI-NEXT: movl $-1, %ebx |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shll %cl, %ebx |
| ; X64-NOBMI-NEXT: notl %ebx |
| ; X64-NOBMI-NEXT: andl %edi, %ebx |
| ; X64-NOBMI-NEXT: movl %esi, %edi |
| ; X64-NOBMI-NEXT: callq use32 |
| ; X64-NOBMI-NEXT: movl %ebx, %eax |
| ; X64-NOBMI-NEXT: popq %rbx |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: pushq %rbx |
| ; X64-BMI1NOTBM-NEXT: shll $8, %edx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orl %edx, %eax |
| ; X64-BMI1NOTBM-NEXT: bextrl %eax, %edi, %ebx |
| ; X64-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X64-BMI1NOTBM-NEXT: callq use32 |
| ; X64-BMI1NOTBM-NEXT: movl %ebx, %eax |
| ; X64-BMI1NOTBM-NEXT: popq %rbx |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr32_b5_skipextrauses: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: pushq %rbx |
| ; X64-BMI1BMI2-NEXT: shrxl %esi, %edi, %eax |
| ; X64-BMI1BMI2-NEXT: bzhil %edx, %eax, %ebx |
| ; X64-BMI1BMI2-NEXT: movl %esi, %edi |
| ; X64-BMI1BMI2-NEXT: callq use32 |
| ; X64-BMI1BMI2-NEXT: movl %ebx, %eax |
| ; X64-BMI1BMI2-NEXT: popq %rbx |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i32 %val, %numskipbits |
| %notmask = shl i32 -1, %numlowbits |
| %mask = xor i32 %notmask, -1 |
| %masked = and i32 %mask, %shifted |
| call void @use32(i32 %numskipbits) |
| ret i32 %masked |
| } |
| |
| ; 64-bit |
| |
| define i64 @bextr64_b0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB23_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB23_2: |
| ; X86-NOBMI-NEXT: movl $-1, %edx |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: shldl %cl, %edx, %edx |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB23_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB23_4: |
| ; X86-NOBMI-NEXT: notl %edx |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB23_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB23_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB23_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: .LBB23_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edx, %esi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edx, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB23_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB23_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %edi, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB23_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB23_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %notmask = shl i64 -1, %numlowbits |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_b1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b1_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB24_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB24_2: |
| ; X86-NOBMI-NEXT: movl $-1, %edx |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: shldl %cl, %edx, %edx |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB24_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB24_4: |
| ; X86-NOBMI-NEXT: notl %edx |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b1_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB24_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB24_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB24_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: .LBB24_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b1_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edx, %esi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edx, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB24_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB24_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %edi, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB24_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB24_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b1_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b1_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b1_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %skip = zext i8 %numskipbits to i64 |
| %shifted = lshr i64 %val, %skip |
| %conv = zext i8 %numlowbits to i64 |
| %notmask = shl i64 -1, %conv |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_b2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b2_load: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: movl 4(%eax), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB25_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB25_2: |
| ; X86-NOBMI-NEXT: movl $-1, %edx |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: shldl %cl, %edx, %edx |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB25_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB25_4: |
| ; X86-NOBMI-NEXT: notl %edx |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b2_load: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1NOTBM-NEXT: movl (%edx), %esi |
| ; X86-BMI1NOTBM-NEXT: movl 4(%edx), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB25_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB25_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB25_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: .LBB25_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b2_load: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl (%edx), %esi |
| ; X86-BMI1BMI2-NEXT: movl 4(%edx), %edi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB25_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB25_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %edi, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB25_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB25_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b2_load: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: movq (%rdi), %rsi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rsi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notq %rax |
| ; X64-NOBMI-NEXT: andq %rsi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b2_load: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, (%rdi), %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b2_load: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, (%rdi), %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i64, i64* %w |
| %shifted = lshr i64 %val, %numskipbits |
| %notmask = shl i64 -1, %numlowbits |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_b3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b3_load_indexzext: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl (%eax), %esi |
| ; X86-NOBMI-NEXT: movl 4(%eax), %eax |
| ; X86-NOBMI-NEXT: movl %eax, %edi |
| ; X86-NOBMI-NEXT: shrl %cl, %edi |
| ; X86-NOBMI-NEXT: shrdl %cl, %eax, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB26_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB26_2: |
| ; X86-NOBMI-NEXT: movl $-1, %edx |
| ; X86-NOBMI-NEXT: movl $-1, %eax |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shll %cl, %eax |
| ; X86-NOBMI-NEXT: shldl %cl, %edx, %edx |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB26_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %eax, %edx |
| ; X86-NOBMI-NEXT: xorl %eax, %eax |
| ; X86-NOBMI-NEXT: .LBB26_4: |
| ; X86-NOBMI-NEXT: notl %edx |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: notl %eax |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1NOTBM-NEXT: movl (%edx), %esi |
| ; X86-BMI1NOTBM-NEXT: movl 4(%edx), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB26_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB26_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB26_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: .LBB26_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b3_load_indexzext: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl (%edx), %esi |
| ; X86-BMI1BMI2-NEXT: movl 4(%edx), %edi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB26_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB26_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %edi, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB26_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB26_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b3_load_indexzext: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: movq (%rdi), %rsi |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rsi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notq %rax |
| ; X64-NOBMI-NEXT: andq %rsi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, (%rdi), %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b3_load_indexzext: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1BMI2-NEXT: # kill: def $esi killed $esi def $rsi |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, (%rdi), %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %val = load i64, i64* %w |
| %skip = zext i8 %numskipbits to i64 |
| %shifted = lshr i64 %val, %skip |
| %conv = zext i8 %numlowbits to i64 |
| %notmask = shl i64 -1, %conv |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %mask, %shifted |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_b4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b4_commutative: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %ch |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl %esi, %edx |
| ; X86-NOBMI-NEXT: shrl %cl, %edx |
| ; X86-NOBMI-NEXT: shrdl %cl, %esi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: je .LBB27_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %edx, %eax |
| ; X86-NOBMI-NEXT: xorl %edx, %edx |
| ; X86-NOBMI-NEXT: .LBB27_2: |
| ; X86-NOBMI-NEXT: movl $-1, %edi |
| ; X86-NOBMI-NEXT: movl $-1, %esi |
| ; X86-NOBMI-NEXT: movb %ch, %cl |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: shldl %cl, %edi, %edi |
| ; X86-NOBMI-NEXT: testb $32, %ch |
| ; X86-NOBMI-NEXT: je .LBB27_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %esi, %edi |
| ; X86-NOBMI-NEXT: xorl %esi, %esi |
| ; X86-NOBMI-NEXT: .LBB27_4: |
| ; X86-NOBMI-NEXT: notl %edi |
| ; X86-NOBMI-NEXT: andl %edi, %edx |
| ; X86-NOBMI-NEXT: notl %esi |
| ; X86-NOBMI-NEXT: andl %esi, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b4_commutative: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: je .LBB27_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %edx, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB27_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %edi |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebx |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB27_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: .LBB27_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b4_commutative: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %edx, %esi |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %edx, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB27_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %edx, %esi |
| ; X86-BMI1BMI2-NEXT: xorl %edx, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB27_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %edi |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %edi, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %edi, %edi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB27_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB27_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %edi, %edx |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b4_commutative: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notq %rax |
| ; X64-NOBMI-NEXT: andq %rdi, %rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b4_commutative: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b4_commutative: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %notmask = shl i64 -1, %numlowbits |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %shifted, %mask ; swapped order |
| ret i64 %masked |
| } |
| |
| define i64 @bextr64_b5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_b5_skipextrauses: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %ebp |
| ; X86-NOBMI-NEXT: pushl %ebx |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: subl $12, %esp |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %ebx |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl %esi, %ebp |
| ; X86-NOBMI-NEXT: movl %eax, %ecx |
| ; X86-NOBMI-NEXT: shrl %cl, %ebp |
| ; X86-NOBMI-NEXT: shrdl %cl, %esi, %ebx |
| ; X86-NOBMI-NEXT: testb $32, %al |
| ; X86-NOBMI-NEXT: je .LBB28_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %ebp, %ebx |
| ; X86-NOBMI-NEXT: xorl %ebp, %ebp |
| ; X86-NOBMI-NEXT: .LBB28_2: |
| ; X86-NOBMI-NEXT: movl $-1, %esi |
| ; X86-NOBMI-NEXT: movl $-1, %edi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %edi |
| ; X86-NOBMI-NEXT: shldl %cl, %esi, %esi |
| ; X86-NOBMI-NEXT: testb $32, %dl |
| ; X86-NOBMI-NEXT: je .LBB28_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: xorl %edi, %edi |
| ; X86-NOBMI-NEXT: .LBB28_4: |
| ; X86-NOBMI-NEXT: notl %esi |
| ; X86-NOBMI-NEXT: andl %ebp, %esi |
| ; X86-NOBMI-NEXT: notl %edi |
| ; X86-NOBMI-NEXT: andl %ebx, %edi |
| ; X86-NOBMI-NEXT: subl $8, %esp |
| ; X86-NOBMI-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-NOBMI-NEXT: pushl %eax |
| ; X86-NOBMI-NEXT: calll use64 |
| ; X86-NOBMI-NEXT: addl $16, %esp |
| ; X86-NOBMI-NEXT: movl %edi, %eax |
| ; X86-NOBMI-NEXT: movl %esi, %edx |
| ; X86-NOBMI-NEXT: addl $12, %esp |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: popl %ebx |
| ; X86-NOBMI-NEXT: popl %ebp |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %ebp |
| ; X86-BMI1NOTBM-NEXT: pushl %ebx |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: subl $12, %esp |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %ebx |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1NOTBM-NEXT: movl %ebx, %esi |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %ebx, %edi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: je .LBB28_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edi |
| ; X86-BMI1NOTBM-NEXT: xorl %esi, %esi |
| ; X86-BMI1NOTBM-NEXT: .LBB28_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebx |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %ebp |
| ; X86-BMI1NOTBM-NEXT: movl %edx, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %ebp |
| ; X86-BMI1NOTBM-NEXT: shldl %cl, %ebx, %ebx |
| ; X86-BMI1NOTBM-NEXT: testb $32, %dl |
| ; X86-BMI1NOTBM-NEXT: je .LBB28_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %ebp, %ebx |
| ; X86-BMI1NOTBM-NEXT: xorl %ebp, %ebp |
| ; X86-BMI1NOTBM-NEXT: .LBB28_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %esi, %ebx, %esi |
| ; X86-BMI1NOTBM-NEXT: andnl %edi, %ebp, %edi |
| ; X86-BMI1NOTBM-NEXT: subl $8, %esp |
| ; X86-BMI1NOTBM-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-BMI1NOTBM-NEXT: pushl %eax |
| ; X86-BMI1NOTBM-NEXT: calll use64 |
| ; X86-BMI1NOTBM-NEXT: addl $16, %esp |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %eax |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: addl $12, %esp |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: popl %ebx |
| ; X86-BMI1NOTBM-NEXT: popl %ebp |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_b5_skipextrauses: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %ebp |
| ; X86-BMI1BMI2-NEXT: pushl %ebx |
| ; X86-BMI1BMI2-NEXT: pushl %edi |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: subl $12, %esp |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-BMI1BMI2-NEXT: movl %eax, %ecx |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %edi |
| ; X86-BMI1BMI2-NEXT: shrxl %eax, %esi, %esi |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: je .LBB28_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: movl %esi, %edi |
| ; X86-BMI1BMI2-NEXT: xorl %esi, %esi |
| ; X86-BMI1BMI2-NEXT: .LBB28_2: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %ebp |
| ; X86-BMI1BMI2-NEXT: shlxl %edx, %ebp, %ebx |
| ; X86-BMI1BMI2-NEXT: movl %edx, %ecx |
| ; X86-BMI1BMI2-NEXT: shldl %cl, %ebp, %ebp |
| ; X86-BMI1BMI2-NEXT: testb $32, %dl |
| ; X86-BMI1BMI2-NEXT: je .LBB28_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl %ebx, %ebp |
| ; X86-BMI1BMI2-NEXT: xorl %ebx, %ebx |
| ; X86-BMI1BMI2-NEXT: .LBB28_4: |
| ; X86-BMI1BMI2-NEXT: andnl %esi, %ebp, %esi |
| ; X86-BMI1BMI2-NEXT: andnl %edi, %ebx, %edi |
| ; X86-BMI1BMI2-NEXT: subl $8, %esp |
| ; X86-BMI1BMI2-NEXT: pushl {{[0-9]+}}(%esp) |
| ; X86-BMI1BMI2-NEXT: pushl %eax |
| ; X86-BMI1BMI2-NEXT: calll use64 |
| ; X86-BMI1BMI2-NEXT: addl $16, %esp |
| ; X86-BMI1BMI2-NEXT: movl %edi, %eax |
| ; X86-BMI1BMI2-NEXT: movl %esi, %edx |
| ; X86-BMI1BMI2-NEXT: addl $12, %esp |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: popl %edi |
| ; X86-BMI1BMI2-NEXT: popl %ebx |
| ; X86-BMI1BMI2-NEXT: popl %ebp |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_b5_skipextrauses: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: pushq %rbx |
| ; X64-NOBMI-NEXT: movl %esi, %ecx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movq $-1, %rbx |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rbx |
| ; X64-NOBMI-NEXT: notq %rbx |
| ; X64-NOBMI-NEXT: andq %rdi, %rbx |
| ; X64-NOBMI-NEXT: movq %rsi, %rdi |
| ; X64-NOBMI-NEXT: callq use64 |
| ; X64-NOBMI-NEXT: movq %rbx, %rax |
| ; X64-NOBMI-NEXT: popq %rbx |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: pushq %rbx |
| ; X64-BMI1NOTBM-NEXT: shlq $8, %rdx |
| ; X64-BMI1NOTBM-NEXT: movzbl %sil, %eax |
| ; X64-BMI1NOTBM-NEXT: orq %rdx, %rax |
| ; X64-BMI1NOTBM-NEXT: bextrq %rax, %rdi, %rbx |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rdi |
| ; X64-BMI1NOTBM-NEXT: callq use64 |
| ; X64-BMI1NOTBM-NEXT: movq %rbx, %rax |
| ; X64-BMI1NOTBM-NEXT: popq %rbx |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_b5_skipextrauses: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: pushq %rbx |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: bzhiq %rdx, %rax, %rbx |
| ; X64-BMI1BMI2-NEXT: movq %rsi, %rdi |
| ; X64-BMI1BMI2-NEXT: callq use64 |
| ; X64-BMI1BMI2-NEXT: movq %rbx, %rax |
| ; X64-BMI1BMI2-NEXT: popq %rbx |
| ; X64-BMI1BMI2-NEXT: retq |
| %shifted = lshr i64 %val, %numskipbits |
| %notmask = shl i64 -1, %numlowbits |
| %mask = xor i64 %notmask, -1 |
| %masked = and i64 %mask, %shifted |
| call void @use64(i64 %numskipbits) |
| ret i64 %masked |
| } |
| |
| ; 64-bit, but with 32-bit output |
| |
| ; Everything done in 64-bit, truncation happens last. |
| define i32 @bextr64_32_b0(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_32_b0: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl %edi, %eax |
| ; X86-NOBMI-NEXT: shrl %cl, %eax |
| ; X86-NOBMI-NEXT: shrdl %cl, %edi, %esi |
| ; X86-NOBMI-NEXT: testb $32, %cl |
| ; X86-NOBMI-NEXT: jne .LBB29_2 |
| ; X86-NOBMI-NEXT: # %bb.1: |
| ; X86-NOBMI-NEXT: movl %esi, %eax |
| ; X86-NOBMI-NEXT: .LBB29_2: |
| ; X86-NOBMI-NEXT: movl $-1, %esi |
| ; X86-NOBMI-NEXT: movl %edx, %ecx |
| ; X86-NOBMI-NEXT: shll %cl, %esi |
| ; X86-NOBMI-NEXT: xorl %ecx, %ecx |
| ; X86-NOBMI-NEXT: testb $32, %dl |
| ; X86-NOBMI-NEXT: jne .LBB29_4 |
| ; X86-NOBMI-NEXT: # %bb.3: |
| ; X86-NOBMI-NEXT: movl %esi, %ecx |
| ; X86-NOBMI-NEXT: .LBB29_4: |
| ; X86-NOBMI-NEXT: notl %ecx |
| ; X86-NOBMI-NEXT: andl %ecx, %eax |
| ; X86-NOBMI-NEXT: popl %esi |
| ; X86-NOBMI-NEXT: popl %edi |
| ; X86-NOBMI-NEXT: retl |
| ; |
| ; X86-BMI1NOTBM-LABEL: bextr64_32_b0: |
| ; X86-BMI1NOTBM: # %bb.0: |
| ; X86-BMI1NOTBM-NEXT: pushl %edi |
| ; X86-BMI1NOTBM-NEXT: pushl %esi |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1NOTBM-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1NOTBM-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-BMI1NOTBM-NEXT: movl %edi, %edx |
| ; X86-BMI1NOTBM-NEXT: shrl %cl, %edx |
| ; X86-BMI1NOTBM-NEXT: shrdl %cl, %edi, %esi |
| ; X86-BMI1NOTBM-NEXT: testb $32, %cl |
| ; X86-BMI1NOTBM-NEXT: jne .LBB29_2 |
| ; X86-BMI1NOTBM-NEXT: # %bb.1: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %edx |
| ; X86-BMI1NOTBM-NEXT: .LBB29_2: |
| ; X86-BMI1NOTBM-NEXT: movl $-1, %esi |
| ; X86-BMI1NOTBM-NEXT: movl %eax, %ecx |
| ; X86-BMI1NOTBM-NEXT: shll %cl, %esi |
| ; X86-BMI1NOTBM-NEXT: xorl %ecx, %ecx |
| ; X86-BMI1NOTBM-NEXT: testb $32, %al |
| ; X86-BMI1NOTBM-NEXT: jne .LBB29_4 |
| ; X86-BMI1NOTBM-NEXT: # %bb.3: |
| ; X86-BMI1NOTBM-NEXT: movl %esi, %ecx |
| ; X86-BMI1NOTBM-NEXT: .LBB29_4: |
| ; X86-BMI1NOTBM-NEXT: andnl %edx, %ecx, %eax |
| ; X86-BMI1NOTBM-NEXT: popl %esi |
| ; X86-BMI1NOTBM-NEXT: popl %edi |
| ; X86-BMI1NOTBM-NEXT: retl |
| ; |
| ; X86-BMI1BMI2-LABEL: bextr64_32_b0: |
| ; X86-BMI1BMI2: # %bb.0: |
| ; X86-BMI1BMI2-NEXT: pushl %esi |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %al |
| ; X86-BMI1BMI2-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %edx |
| ; X86-BMI1BMI2-NEXT: movl {{[0-9]+}}(%esp), %esi |
| ; X86-BMI1BMI2-NEXT: shrdl %cl, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: testb $32, %cl |
| ; X86-BMI1BMI2-NEXT: je .LBB29_2 |
| ; X86-BMI1BMI2-NEXT: # %bb.1: |
| ; X86-BMI1BMI2-NEXT: shrxl %ecx, %esi, %edx |
| ; X86-BMI1BMI2-NEXT: .LBB29_2: |
| ; X86-BMI1BMI2-NEXT: xorl %ecx, %ecx |
| ; X86-BMI1BMI2-NEXT: testb $32, %al |
| ; X86-BMI1BMI2-NEXT: jne .LBB29_4 |
| ; X86-BMI1BMI2-NEXT: # %bb.3: |
| ; X86-BMI1BMI2-NEXT: movl $-1, %ecx |
| ; X86-BMI1BMI2-NEXT: shlxl %eax, %ecx, %ecx |
| ; X86-BMI1BMI2-NEXT: .LBB29_4: |
| ; X86-BMI1BMI2-NEXT: andnl %edx, %ecx, %eax |
| ; X86-BMI1BMI2-NEXT: popl %esi |
| ; X86-BMI1BMI2-NEXT: retl |
| ; |
| ; X64-NOBMI-LABEL: bextr64_32_b0: |
| ; X64-NOBMI: # %bb.0: |
| ; X64-NOBMI-NEXT: movq %rsi, %rcx |
| ; X64-NOBMI-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-NOBMI-NEXT: shrq %cl, %rdi |
| ; X64-NOBMI-NEXT: movq $-1, %rax |
| ; X64-NOBMI-NEXT: movl %edx, %ecx |
| ; X64-NOBMI-NEXT: shlq %cl, %rax |
| ; X64-NOBMI-NEXT: notl %eax |
| ; X64-NOBMI-NEXT: andl %edi, %eax |
| ; X64-NOBMI-NEXT: # kill: def $eax killed $eax killed $rax |
| ; X64-NOBMI-NEXT: retq |
| ; |
| ; X64-BMI1NOTBM-LABEL: bextr64_32_b0: |
| ; X64-BMI1NOTBM: # %bb.0: |
| ; X64-BMI1NOTBM-NEXT: movq %rsi, %rcx |
| ; X64-BMI1NOTBM-NEXT: # kill: def $cl killed $cl killed $rcx |
| ; X64-BMI1NOTBM-NEXT: shrq %cl, %rdi |
| ; X64-BMI1NOTBM-NEXT: movq $-1, %rax |
| ; X64-BMI1NOTBM-NEXT: movl %edx, %ecx |
| ; X64-BMI1NOTBM-NEXT: shlq %cl, %rax |
| ; X64-BMI1NOTBM-NEXT: andnl %edi, %eax, %eax |
| ; X64-BMI1NOTBM-NEXT: retq |
| ; |
| ; X64-BMI1BMI2-LABEL: bextr64_32_b0: |
| ; X64-BMI1BMI2: # %bb.0: |
| ; X64-BMI1BMI2-NEXT: # kill: def $edx killed $edx def $rdx |
| ; X64-BMI1BMI2-NEXT: shrxq %rsi, %rdi, %rax |
| ; X64-BMI1BMI2-NEXT: movq $-1, %rcx |
| ; X64-BMI1BMI2-NEXT: shlxq %rdx, %rcx, %rcx |
| ; X64-BMI1BMI2-NEXT: andnl %eax, %ecx, %eax |
| ; X64-BMI1BMI2-NEXT: retq |
| %shiftedval = lshr i64 %val, %numskipbits |
| %widenumlowbits = zext i8 %numlowbits to i64 |
| %notmask = shl nsw i64 -1, %widenumlowbits |
| %mask = xor i64 %notmask, -1 |
| %wideres = and i64 %shiftedval, %mask |
| %res = trunc i64 %wideres to i32 |
| ret i32 %res |
| } |
| |
| ; Shifting happens in 64-bit, then truncation. Masking is 32-bit. |
| define i32 @bextr64_32_b1(i64 %val, i64 %numskipbits, i8 %numlowbits) nounwind { |
| ; X86-NOBMI-LABEL: bextr64_32_b1: |
| ; X86-NOBMI: # %bb.0: |
| ; X86-NOBMI-NEXT: pushl %edi |
| ; X86-NOBMI-NEXT: pushl %esi |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %dl |
| ; X86-NOBMI-NEXT: movb {{[0-9]+}}(%esp), %cl |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %eax |
| ; X86-NOBMI-NEXT: movl {{[0-9]+}}(%esp), %edi |
| ; X86-NOBMI-NEXT: movl %edi, %esi |
| ; X86-NOBMI-NEXT: shrl %cl, %esi |
| ; X86-NOBMI-NEXT: shrdl %cl, %edi, %eax |
| ; X86-NOBMI-NEXT: testb $32, %cl |
|