[llvm] r317698 - [X86] Add some initial scheduling tests for generic x86 instructions
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Wed Nov 8 08:35:42 PST 2017
Author: rksimon
Date: Wed Nov 8 08:35:42 2017
New Revision: 317698
URL: http://llvm.org/viewvc/llvm-project?rev=317698&view=rev
Log:
[X86] Add some initial scheduling tests for generic x86 instructions
These will be using inline asm to ensure we have coverage that we're unlikely to get from lowering of basic ir.
Currently waiting for D39728 to land to add support for scheduler comments for inline asm.
Added:
llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll
llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll
Added: llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll?rev=317698&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll (added)
+++ llvm/trunk/test/CodeGen/X86/schedule-x86_32.ll Wed Nov 8 08:35:42 2017
@@ -0,0 +1,348 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=i686 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=atom | FileCheck %s --check-prefix=CHECK --check-prefix=ATOM
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=slm | FileCheck %s --check-prefix=CHECK --check-prefix=SLM
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=sandybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=ivybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
+; RUN: llc < %s -mtriple=i686-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
+
+define i8 @test_aaa(i8 %a0) optsize {
+; GENERIC-LABEL: test_aaa:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movb {{[0-9]+}}(%esp), %al
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: aaa
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: retl
+;
+; ATOM-LABEL: test_aaa:
+; ATOM: # BB#0:
+; ATOM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:1.00]
+; ATOM-NEXT: #APP
+; ATOM-NEXT: aaa
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: retl # sched: [79:39.50]
+;
+; SLM-LABEL: test_aaa:
+; SLM: # BB#0:
+; SLM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [3:1.00]
+; SLM-NEXT: #APP
+; SLM-NEXT: aaa
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: retl # sched: [4:1.00]
+;
+; SANDY-LABEL: test_aaa:
+; SANDY: # BB#0:
+; SANDY-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SANDY-NEXT: #APP
+; SANDY-NEXT: aaa
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: retl # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_aaa:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:0.50]
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: aaa
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: retl # sched: [5:0.50]
+;
+; BROADWELL-LABEL: test_aaa:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: aaa
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: retl # sched: [6:0.50]
+;
+; SKYLAKE-LABEL: test_aaa:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: aaa
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: retl # sched: [6:0.50]
+;
+; SKX-LABEL: test_aaa:
+; SKX: # BB#0:
+; SKX-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKX-NEXT: #APP
+; SKX-NEXT: aaa
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: retl # sched: [6:0.50]
+;
+; BTVER2-LABEL: test_aaa:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:1.00]
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: aaa
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: retl # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_aaa:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [8:0.50]
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: aaa
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: retl # sched: [1:0.50]
+ %1 = tail call i8 asm "aaa", "=r,r"(i8 %a0) nounwind
+ ret i8 %1
+}
+
+define i8 @test_aad(i16 %a0) optsize {
+; GENERIC-LABEL: test_aad:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: aad
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: retl
+;
+; ATOM-LABEL: test_aad:
+; ATOM: # BB#0:
+; ATOM-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [1:1.00]
+; ATOM-NEXT: #APP
+; ATOM-NEXT: aad
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: retl # sched: [79:39.50]
+;
+; SLM-LABEL: test_aad:
+; SLM: # BB#0:
+; SLM-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [4:1.00]
+; SLM-NEXT: #APP
+; SLM-NEXT: aad
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: retl # sched: [4:1.00]
+;
+; SANDY-LABEL: test_aad:
+; SANDY: # BB#0:
+; SANDY-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
+; SANDY-NEXT: #APP
+; SANDY-NEXT: aad
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: retl # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_aad:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [4:0.50]
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: aad
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: retl # sched: [5:0.50]
+;
+; BROADWELL-LABEL: test_aad:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: aad
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: retl # sched: [6:0.50]
+;
+; SKYLAKE-LABEL: test_aad:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: aad
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: retl # sched: [6:0.50]
+;
+; SKX-LABEL: test_aad:
+; SKX: # BB#0:
+; SKX-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [5:0.50]
+; SKX-NEXT: #APP
+; SKX-NEXT: aad
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: retl # sched: [6:0.50]
+;
+; BTVER2-LABEL: test_aad:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [4:1.00]
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: aad
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: retl # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_aad:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: movzwl {{[0-9]+}}(%esp), %eax # sched: [8:0.50]
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: aad
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: retl # sched: [1:0.50]
+ %1 = tail call i8 asm "aad", "=r,r"(i16 %a0) nounwind
+ ret i8 %1
+}
+
+define i16 @test_aam(i8 %a0) optsize {
+; GENERIC-LABEL: test_aam:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movb {{[0-9]+}}(%esp), %al
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: aam
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: retl
+;
+; ATOM-LABEL: test_aam:
+; ATOM: # BB#0:
+; ATOM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:1.00]
+; ATOM-NEXT: #APP
+; ATOM-NEXT: aam
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: retl # sched: [79:39.50]
+;
+; SLM-LABEL: test_aam:
+; SLM: # BB#0:
+; SLM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [3:1.00]
+; SLM-NEXT: #APP
+; SLM-NEXT: aam
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: retl # sched: [4:1.00]
+;
+; SANDY-LABEL: test_aam:
+; SANDY: # BB#0:
+; SANDY-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SANDY-NEXT: #APP
+; SANDY-NEXT: aam
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: retl # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_aam:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:0.50]
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: aam
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: retl # sched: [5:0.50]
+;
+; BROADWELL-LABEL: test_aam:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: aam
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: retl # sched: [6:0.50]
+;
+; SKYLAKE-LABEL: test_aam:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: aam
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: retl # sched: [6:0.50]
+;
+; SKX-LABEL: test_aam:
+; SKX: # BB#0:
+; SKX-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKX-NEXT: #APP
+; SKX-NEXT: aam
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: retl # sched: [6:0.50]
+;
+; BTVER2-LABEL: test_aam:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:1.00]
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: aam
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: retl # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_aam:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [8:0.50]
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: aam
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: retl # sched: [1:0.50]
+ %1 = tail call i16 asm "aam", "=r,r"(i8 %a0) nounwind
+ ret i16 %1
+}
+
+define i8 @test_aas(i8 %a0) optsize {
+; GENERIC-LABEL: test_aas:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: movb {{[0-9]+}}(%esp), %al
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: aas
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: retl
+;
+; ATOM-LABEL: test_aas:
+; ATOM: # BB#0:
+; ATOM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:1.00]
+; ATOM-NEXT: #APP
+; ATOM-NEXT: aas
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: retl # sched: [79:39.50]
+;
+; SLM-LABEL: test_aas:
+; SLM: # BB#0:
+; SLM-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [3:1.00]
+; SLM-NEXT: #APP
+; SLM-NEXT: aas
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: retl # sched: [4:1.00]
+;
+; SANDY-LABEL: test_aas:
+; SANDY: # BB#0:
+; SANDY-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SANDY-NEXT: #APP
+; SANDY-NEXT: aas
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: retl # sched: [5:1.00]
+;
+; HASWELL-LABEL: test_aas:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [1:0.50]
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: aas
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: retl # sched: [5:0.50]
+;
+; BROADWELL-LABEL: test_aas:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: aas
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: retl # sched: [6:0.50]
+;
+; SKYLAKE-LABEL: test_aas:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: aas
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: retl # sched: [6:0.50]
+;
+; SKX-LABEL: test_aas:
+; SKX: # BB#0:
+; SKX-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:0.50]
+; SKX-NEXT: #APP
+; SKX-NEXT: aas
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: retl # sched: [6:0.50]
+;
+; BTVER2-LABEL: test_aas:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [5:1.00]
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: aas
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: retl # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_aas:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: movb {{[0-9]+}}(%esp), %al # sched: [8:0.50]
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: aas
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: retl # sched: [1:0.50]
+ %1 = tail call i8 asm "aas", "=r,r"(i8 %a0) nounwind
+ ret i8 %1
+}
Added: llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll?rev=317698&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll (added)
+++ llvm/trunk/test/CodeGen/X86/schedule-x86_64.ll Wed Nov 8 08:35:42 2017
@@ -0,0 +1,737 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=atom | FileCheck %s --check-prefix=CHECK --check-prefix=ATOM
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=slm | FileCheck %s --check-prefix=CHECK --check-prefix=SLM
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
+
+define i16 @test_bsf16(i16 %a0, i16* %a1) optsize {
+; GENERIC-LABEL: test_bsf16:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsfw %di, %ax
+; GENERIC-NEXT: bsfw (%rsi), %cx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; GENERIC-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsf16:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsfw %di, %ax
+; ATOM-NEXT: bsfw (%rsi), %cx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; ATOM-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsf16:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsfw %di, %ax
+; SLM-NEXT: bsfw (%rsi), %cx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; SLM-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsf16:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsfw %di, %ax
+; SANDY-NEXT: bsfw (%rsi), %cx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; SANDY-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsf16:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsfw %di, %ax
+; HASWELL-NEXT: bsfw (%rsi), %cx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsf16:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsfw %di, %ax
+; BROADWELL-NEXT: bsfw (%rsi), %cx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsf16:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsfw %di, %ax
+; SKYLAKE-NEXT: bsfw (%rsi), %cx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKYLAKE-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsf16:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsfw %di, %ax
+; SKX-NEXT: bsfw (%rsi), %cx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKX-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsf16:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsfw %di, %ax
+; BTVER2-NEXT: bsfw (%rsi), %cx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; BTVER2-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsf16:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsfw %di, %ax
+; ZNVER1-NEXT: bsfw (%rsi), %cx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; ZNVER1-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i16, i16 } asm sideeffect "bsf $2, $0 \0A\09 bsf $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i16 %a0, i16* %a1)
+ %2 = extractvalue { i16, i16 } %1, 0
+ %3 = extractvalue { i16, i16 } %1, 1
+ %4 = or i16 %2, %3
+ ret i16 %4
+}
+define i32 @test_bsf32(i32 %a0, i32* %a1) optsize {
+; GENERIC-LABEL: test_bsf32:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsfl %edi, %eax
+; GENERIC-NEXT: bsfl (%rsi), %ecx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsf32:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsfl %edi, %eax
+; ATOM-NEXT: bsfl (%rsi), %ecx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsf32:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsfl %edi, %eax
+; SLM-NEXT: bsfl (%rsi), %ecx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsf32:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsfl %edi, %eax
+; SANDY-NEXT: bsfl (%rsi), %ecx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsf32:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsfl %edi, %eax
+; HASWELL-NEXT: bsfl (%rsi), %ecx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsf32:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsfl %edi, %eax
+; BROADWELL-NEXT: bsfl (%rsi), %ecx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsf32:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsfl %edi, %eax
+; SKYLAKE-NEXT: bsfl (%rsi), %ecx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsf32:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsfl %edi, %eax
+; SKX-NEXT: bsfl (%rsi), %ecx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsf32:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsfl %edi, %eax
+; BTVER2-NEXT: bsfl (%rsi), %ecx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsf32:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsfl %edi, %eax
+; ZNVER1-NEXT: bsfl (%rsi), %ecx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i32, i32 } asm sideeffect "bsf $2, $0 \0A\09 bsf $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i32 %a0, i32* %a1)
+ %2 = extractvalue { i32, i32 } %1, 0
+ %3 = extractvalue { i32, i32 } %1, 1
+ %4 = or i32 %2, %3
+ ret i32 %4
+}
+define i64 @test_bsf64(i64 %a0, i64* %a1) optsize {
+; GENERIC-LABEL: test_bsf64:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsfq %rdi, %rax
+; GENERIC-NEXT: bsfq (%rsi), %rcx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orq %rcx, %rax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsf64:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsfq %rdi, %rax
+; ATOM-NEXT: bsfq (%rsi), %rcx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsf64:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsfq %rdi, %rax
+; SLM-NEXT: bsfq (%rsi), %rcx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsf64:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsfq %rdi, %rax
+; SANDY-NEXT: bsfq (%rsi), %rcx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orq %rcx, %rax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsf64:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsfq %rdi, %rax
+; HASWELL-NEXT: bsfq (%rsi), %rcx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsf64:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsfq %rdi, %rax
+; BROADWELL-NEXT: bsfq (%rsi), %rcx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsf64:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsfq %rdi, %rax
+; SKYLAKE-NEXT: bsfq (%rsi), %rcx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsf64:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsfq %rdi, %rax
+; SKX-NEXT: bsfq (%rsi), %rcx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsf64:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsfq %rdi, %rax
+; BTVER2-NEXT: bsfq (%rsi), %rcx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsf64:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsfq %rdi, %rax
+; ZNVER1-NEXT: bsfq (%rsi), %rcx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i64, i64 } asm sideeffect "bsf $2, $0 \0A\09 bsf $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i64 %a0, i64* %a1)
+ %2 = extractvalue { i64, i64 } %1, 0
+ %3 = extractvalue { i64, i64 } %1, 1
+ %4 = or i64 %2, %3
+ ret i64 %4
+}
+
+define i16 @test_bsr16(i16 %a0, i16* %a1) optsize {
+; GENERIC-LABEL: test_bsr16:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsrw %di, %ax
+; GENERIC-NEXT: bsrw (%rsi), %cx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; GENERIC-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsr16:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsrw %di, %ax
+; ATOM-NEXT: bsrw (%rsi), %cx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; ATOM-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsr16:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsrw %di, %ax
+; SLM-NEXT: bsrw (%rsi), %cx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; SLM-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsr16:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsrw %di, %ax
+; SANDY-NEXT: bsrw (%rsi), %cx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; SANDY-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsr16:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsrw %di, %ax
+; HASWELL-NEXT: bsrw (%rsi), %cx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; HASWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsr16:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsrw %di, %ax
+; BROADWELL-NEXT: bsrw (%rsi), %cx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; BROADWELL-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsr16:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsrw %di, %ax
+; SKYLAKE-NEXT: bsrw (%rsi), %cx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKYLAKE-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsr16:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsrw %di, %ax
+; SKX-NEXT: bsrw (%rsi), %cx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKX-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsr16:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsrw %di, %ax
+; BTVER2-NEXT: bsrw (%rsi), %cx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; BTVER2-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsr16:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsrw %di, %ax
+; ZNVER1-NEXT: bsrw (%rsi), %cx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; ZNVER1-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill>
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i16, i16 } asm sideeffect "bsr $2, $0 \0A\09 bsr $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i16 %a0, i16* %a1)
+ %2 = extractvalue { i16, i16 } %1, 0
+ %3 = extractvalue { i16, i16 } %1, 1
+ %4 = or i16 %2, %3
+ ret i16 %4
+}
+define i32 @test_bsr32(i32 %a0, i32* %a1) optsize {
+; GENERIC-LABEL: test_bsr32:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsrl %edi, %eax
+; GENERIC-NEXT: bsrl (%rsi), %ecx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsr32:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsrl %edi, %eax
+; ATOM-NEXT: bsrl (%rsi), %ecx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsr32:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsrl %edi, %eax
+; SLM-NEXT: bsrl (%rsi), %ecx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsr32:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsrl %edi, %eax
+; SANDY-NEXT: bsrl (%rsi), %ecx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orl %ecx, %eax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsr32:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsrl %edi, %eax
+; HASWELL-NEXT: bsrl (%rsi), %ecx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsr32:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsrl %edi, %eax
+; BROADWELL-NEXT: bsrl (%rsi), %ecx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsr32:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsrl %edi, %eax
+; SKYLAKE-NEXT: bsrl (%rsi), %ecx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsr32:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsrl %edi, %eax
+; SKX-NEXT: bsrl (%rsi), %ecx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsr32:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsrl %edi, %eax
+; BTVER2-NEXT: bsrl (%rsi), %ecx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orl %ecx, %eax # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsr32:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsrl %edi, %eax
+; ZNVER1-NEXT: bsrl (%rsi), %ecx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orl %ecx, %eax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i32, i32 } asm sideeffect "bsr $2, $0 \0A\09 bsr $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i32 %a0, i32* %a1)
+ %2 = extractvalue { i32, i32 } %1, 0
+ %3 = extractvalue { i32, i32 } %1, 1
+ %4 = or i32 %2, %3
+ ret i32 %4
+}
+define i64 @test_bsr64(i64 %a0, i64* %a1) optsize {
+; GENERIC-LABEL: test_bsr64:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: #APP
+; GENERIC-NEXT: bsrq %rdi, %rax
+; GENERIC-NEXT: bsrq (%rsi), %rcx
+; GENERIC-NEXT: #NO_APP
+; GENERIC-NEXT: orq %rcx, %rax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bsr64:
+; ATOM: # BB#0:
+; ATOM-NEXT: #APP
+; ATOM-NEXT: bsrq %rdi, %rax
+; ATOM-NEXT: bsrq (%rsi), %rcx
+; ATOM-NEXT: #NO_APP
+; ATOM-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bsr64:
+; SLM: # BB#0:
+; SLM-NEXT: #APP
+; SLM-NEXT: bsrq %rdi, %rax
+; SLM-NEXT: bsrq (%rsi), %rcx
+; SLM-NEXT: #NO_APP
+; SLM-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bsr64:
+; SANDY: # BB#0:
+; SANDY-NEXT: #APP
+; SANDY-NEXT: bsrq %rdi, %rax
+; SANDY-NEXT: bsrq (%rsi), %rcx
+; SANDY-NEXT: #NO_APP
+; SANDY-NEXT: orq %rcx, %rax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bsr64:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: #APP
+; HASWELL-NEXT: bsrq %rdi, %rax
+; HASWELL-NEXT: bsrq (%rsi), %rcx
+; HASWELL-NEXT: #NO_APP
+; HASWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bsr64:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: #APP
+; BROADWELL-NEXT: bsrq %rdi, %rax
+; BROADWELL-NEXT: bsrq (%rsi), %rcx
+; BROADWELL-NEXT: #NO_APP
+; BROADWELL-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bsr64:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: #APP
+; SKYLAKE-NEXT: bsrq %rdi, %rax
+; SKYLAKE-NEXT: bsrq (%rsi), %rcx
+; SKYLAKE-NEXT: #NO_APP
+; SKYLAKE-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bsr64:
+; SKX: # BB#0:
+; SKX-NEXT: #APP
+; SKX-NEXT: bsrq %rdi, %rax
+; SKX-NEXT: bsrq (%rsi), %rcx
+; SKX-NEXT: #NO_APP
+; SKX-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bsr64:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: #APP
+; BTVER2-NEXT: bsrq %rdi, %rax
+; BTVER2-NEXT: bsrq (%rsi), %rcx
+; BTVER2-NEXT: #NO_APP
+; BTVER2-NEXT: orq %rcx, %rax # sched: [1:0.50]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bsr64:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: #APP
+; ZNVER1-NEXT: bsrq %rdi, %rax
+; ZNVER1-NEXT: bsrq (%rsi), %rcx
+; ZNVER1-NEXT: #NO_APP
+; ZNVER1-NEXT: orq %rcx, %rax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = call { i64, i64 } asm sideeffect "bsr $2, $0 \0A\09 bsr $3, $1", "=r,=r,r,*m,~{dirflag},~{fpsr},~{flags}"(i64 %a0, i64* %a1)
+ %2 = extractvalue { i64, i64 } %1, 0
+ %3 = extractvalue { i64, i64 } %1, 1
+ %4 = or i64 %2, %3
+ ret i64 %4
+}
+
+define i32 @test_bswap32(i32 %a0) optsize {
+; GENERIC-LABEL: test_bswap32:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: bswapl %edi # sched: [2:1.00]
+; GENERIC-NEXT: movl %edi, %eax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bswap32:
+; ATOM: # BB#0:
+; ATOM-NEXT: bswapl %edi # sched: [1:1.00]
+; ATOM-NEXT: movl %edi, %eax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bswap32:
+; SLM: # BB#0:
+; SLM-NEXT: bswapl %edi # sched: [1:0.50]
+; SLM-NEXT: movl %edi, %eax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bswap32:
+; SANDY: # BB#0:
+; SANDY-NEXT: bswapl %edi # sched: [2:1.00]
+; SANDY-NEXT: movl %edi, %eax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bswap32:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: bswapl %edi # sched: [2:0.50]
+; HASWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bswap32:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: bswapl %edi # sched: [2:0.50]
+; BROADWELL-NEXT: movl %edi, %eax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bswap32:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: bswapl %edi # sched: [2:0.50]
+; SKYLAKE-NEXT: movl %edi, %eax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bswap32:
+; SKX: # BB#0:
+; SKX-NEXT: bswapl %edi # sched: [2:0.50]
+; SKX-NEXT: movl %edi, %eax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bswap32:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: bswapl %edi # sched: [1:0.50]
+; BTVER2-NEXT: movl %edi, %eax # sched: [1:0.17]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bswap32:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: bswapl %edi # sched: [1:1.00]
+; ZNVER1-NEXT: movl %edi, %eax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = tail call i32 asm "bswap $0", "=r,0"(i32 %a0) nounwind
+ ret i32 %1
+}
+define i64 @test_bswap64(i64 %a0) optsize {
+; GENERIC-LABEL: test_bswap64:
+; GENERIC: # BB#0:
+; GENERIC-NEXT: bswapq %rdi # sched: [2:1.00]
+; GENERIC-NEXT: movq %rdi, %rax # sched: [1:0.33]
+; GENERIC-NEXT: retq # sched: [1:1.00]
+;
+; ATOM-LABEL: test_bswap64:
+; ATOM: # BB#0:
+; ATOM-NEXT: bswapq %rdi # sched: [1:1.00]
+; ATOM-NEXT: movq %rdi, %rax # sched: [1:0.50]
+; ATOM-NEXT: retq # sched: [79:39.50]
+;
+; SLM-LABEL: test_bswap64:
+; SLM: # BB#0:
+; SLM-NEXT: bswapq %rdi # sched: [1:0.50]
+; SLM-NEXT: movq %rdi, %rax # sched: [1:0.50]
+; SLM-NEXT: retq # sched: [4:1.00]
+;
+; SANDY-LABEL: test_bswap64:
+; SANDY: # BB#0:
+; SANDY-NEXT: bswapq %rdi # sched: [2:1.00]
+; SANDY-NEXT: movq %rdi, %rax # sched: [1:0.33]
+; SANDY-NEXT: retq # sched: [1:1.00]
+;
+; HASWELL-LABEL: test_bswap64:
+; HASWELL: # BB#0:
+; HASWELL-NEXT: bswapq %rdi # sched: [2:0.50]
+; HASWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
+; HASWELL-NEXT: retq # sched: [2:1.00]
+;
+; BROADWELL-LABEL: test_bswap64:
+; BROADWELL: # BB#0:
+; BROADWELL-NEXT: bswapq %rdi # sched: [2:0.50]
+; BROADWELL-NEXT: movq %rdi, %rax # sched: [1:0.25]
+; BROADWELL-NEXT: retq # sched: [7:1.00]
+;
+; SKYLAKE-LABEL: test_bswap64:
+; SKYLAKE: # BB#0:
+; SKYLAKE-NEXT: bswapq %rdi # sched: [2:0.50]
+; SKYLAKE-NEXT: movq %rdi, %rax # sched: [1:0.25]
+; SKYLAKE-NEXT: retq # sched: [7:1.00]
+;
+; SKX-LABEL: test_bswap64:
+; SKX: # BB#0:
+; SKX-NEXT: bswapq %rdi # sched: [2:0.50]
+; SKX-NEXT: movq %rdi, %rax # sched: [1:0.25]
+; SKX-NEXT: retq # sched: [7:1.00]
+;
+; BTVER2-LABEL: test_bswap64:
+; BTVER2: # BB#0:
+; BTVER2-NEXT: bswapq %rdi # sched: [1:0.50]
+; BTVER2-NEXT: movq %rdi, %rax # sched: [1:0.17]
+; BTVER2-NEXT: retq # sched: [4:1.00]
+;
+; ZNVER1-LABEL: test_bswap64:
+; ZNVER1: # BB#0:
+; ZNVER1-NEXT: bswapq %rdi # sched: [1:1.00]
+; ZNVER1-NEXT: movq %rdi, %rax # sched: [1:0.25]
+; ZNVER1-NEXT: retq # sched: [1:0.50]
+ %1 = tail call i64 asm "bswap $0", "=r,0"(i64 %a0) nounwind
+ ret i64 %1
+}
More information about the llvm-commits
mailing list