summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/sse3-schedule.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/sse3-schedule.ll')
-rw-r--r--llvm/test/CodeGen/X86/sse3-schedule.ll562
1 files changed, 551 insertions, 11 deletions
diff --git a/llvm/test/CodeGen/X86/sse3-schedule.ll b/llvm/test/CodeGen/X86/sse3-schedule.ll
index dc23cc6968d..50bdf203c74 100644
--- a/llvm/test/CodeGen/X86/sse3-schedule.ll
+++ b/llvm/test/CodeGen/X86/sse3-schedule.ll
@@ -1,15 +1,23 @@
; 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 -mattr=+sse3 | 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
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,GENERIC
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=atom -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,ATOM
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=slm -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SLM
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SANDY-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SANDY-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SANDY
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,HASWELL-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,HASWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BROADWELL-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BROADWELL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SKYLAKE-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKYLAKE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,SKX-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,SKX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,BTVER2-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,BTVER2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-ssse3 | FileCheck %s --check-prefixes=CHECK,ZNVER1-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 -mattr=-avx2 | FileCheck %s --check-prefixes=CHECK,ZNVER1
define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
; GENERIC-LABEL: test_addsubpd:
@@ -30,42 +38,84 @@ define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x doub
; SLM-NEXT: addsubpd (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_addsubpd:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [9:1.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_addsubpd:
; SANDY: # %bb.0:
; SANDY-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_addsubpd:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [9:1.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_addsubpd:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_addsubpd:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; BROADWELL-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [8:1.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_addsubpd:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_addsubpd:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [4:0.50]
+; SKYLAKE-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [10:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_addsubpd:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
; SKYLAKE-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_addsubpd:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [4:0.33]
+; SKX-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [10:0.50]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_addsubpd:
; SKX: # %bb.0:
; SKX-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
; SKX-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_addsubpd:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_addsubpd:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_addsubpd:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; ZNVER1-SSE-NEXT: addsubpd (%rdi), %xmm0 # sched: [10:1.00]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_addsubpd:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
@@ -97,42 +147,84 @@ define <4 x float> @test_addsubps(<4 x float> %a0, <4 x float> %a1, <4 x float>
; SLM-NEXT: addsubps (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_addsubps:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [9:1.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_addsubps:
; SANDY: # %bb.0:
; SANDY-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_addsubps:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [9:1.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_addsubps:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_addsubps:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
+; BROADWELL-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [8:1.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_addsubps:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_addsubps:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [4:0.50]
+; SKYLAKE-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [10:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_addsubps:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
; SKYLAKE-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_addsubps:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [4:0.33]
+; SKX-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [10:0.50]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_addsubps:
; SKX: # %bb.0:
; SKX-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
; SKX-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_addsubps:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_addsubps:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_addsubps:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
+; ZNVER1-SSE-NEXT: addsubps (%rdi), %xmm0 # sched: [10:1.00]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_addsubps:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
@@ -164,42 +256,84 @@ define <2 x double> @test_haddpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; SLM-NEXT: haddpd (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_haddpd:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [5:2.00]
+; SANDY-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [11:2.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_haddpd:
; SANDY: # %bb.0:
; SANDY-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_haddpd:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [5:2.00]
+; HASWELL-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [11:2.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_haddpd:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_haddpd:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [5:2.00]
+; BROADWELL-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [10:2.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_haddpd:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_haddpd:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [6:2.00]
+; SKYLAKE-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [12:2.00]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_haddpd:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKYLAKE-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_haddpd:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [6:2.00]
+; SKX-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [12:2.00]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_haddpd:
; SKX: # %bb.0:
; SKX-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKX-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_haddpd:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_haddpd:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_haddpd:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: haddpd %xmm1, %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: haddpd (%rdi), %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_haddpd:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [100:?]
@@ -231,42 +365,84 @@ define <4 x float> @test_haddps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; SLM-NEXT: haddps (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_haddps:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [5:2.00]
+; SANDY-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [11:2.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_haddps:
; SANDY: # %bb.0:
; SANDY-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_haddps:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [5:2.00]
+; HASWELL-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [11:2.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_haddps:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_haddps:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [5:2.00]
+; BROADWELL-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [10:2.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_haddps:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_haddps:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [6:2.00]
+; SKYLAKE-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [12:2.00]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_haddps:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKYLAKE-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_haddps:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [6:2.00]
+; SKX-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [12:2.00]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_haddps:
; SKX: # %bb.0:
; SKX-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKX-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_haddps:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_haddps:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_haddps:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: haddps %xmm1, %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: haddps (%rdi), %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_haddps:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [100:?]
@@ -298,42 +474,84 @@ define <2 x double> @test_hsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double
; SLM-NEXT: hsubpd (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_hsubpd:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [5:2.00]
+; SANDY-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [11:2.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_hsubpd:
; SANDY: # %bb.0:
; SANDY-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_hsubpd:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [5:2.00]
+; HASWELL-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [11:2.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_hsubpd:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_hsubpd:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [5:2.00]
+; BROADWELL-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [10:2.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_hsubpd:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_hsubpd:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [6:2.00]
+; SKYLAKE-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [12:2.00]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_hsubpd:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKYLAKE-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_hsubpd:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [6:2.00]
+; SKX-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [12:2.00]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_hsubpd:
; SKX: # %bb.0:
; SKX-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKX-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_hsubpd:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_hsubpd:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_hsubpd:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: hsubpd %xmm1, %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: hsubpd (%rdi), %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_hsubpd:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [100:?]
@@ -365,42 +583,84 @@ define <4 x float> @test_hsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%
; SLM-NEXT: hsubps (%rdi), %xmm0 # sched: [6:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_hsubps:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [5:2.00]
+; SANDY-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [11:2.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_hsubps:
; SANDY: # %bb.0:
; SANDY-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; SANDY-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_hsubps:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [5:2.00]
+; HASWELL-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [11:2.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_hsubps:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; HASWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_hsubps:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [5:2.00]
+; BROADWELL-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [10:2.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_hsubps:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
; BROADWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_hsubps:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [6:2.00]
+; SKYLAKE-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [12:2.00]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_hsubps:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKYLAKE-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_hsubps:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [6:2.00]
+; SKX-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [12:2.00]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_hsubps:
; SKX: # %bb.0:
; SKX-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
; SKX-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_hsubps:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [8:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_hsubps:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_hsubps:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: hsubps %xmm1, %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: hsubps (%rdi), %xmm0 # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_hsubps:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [100:?]
@@ -431,36 +691,71 @@ define <16 x i8> @test_lddqu(i8* %a0) {
; SLM-NEXT: lddqu (%rdi), %xmm0 # sched: [3:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_lddqu:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [6:0.50]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_lddqu:
; SANDY: # %bb.0:
; SANDY-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_lddqu:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [6:0.50]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_lddqu:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_lddqu:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [5:0.50]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_lddqu:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [5:0.50]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_lddqu:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [6:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_lddqu:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_lddqu:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [6:0.50]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_lddqu:
; SKX: # %bb.0:
; SKX-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_lddqu:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [5:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_lddqu:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vlddqu (%rdi), %xmm0 # sched: [5:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_lddqu:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: lddqu (%rdi), %xmm0 # sched: [8:0.50]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_lddqu:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vlddqu (%rdi), %xmm0 # sched: [8:0.50]
@@ -492,6 +787,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; SLM-NEXT: monitor # sched: [100:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_monitor:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; SANDY-SSE-NEXT: movl %esi, %ecx # sched: [1:0.33]
+; SANDY-SSE-NEXT: monitor # sched: [100:0.33]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_monitor:
; SANDY: # %bb.0:
; SANDY-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -499,6 +801,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; SANDY-NEXT: monitor # sched: [100:0.33]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_monitor:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; HASWELL-SSE-NEXT: movl %esi, %ecx # sched: [1:0.25]
+; HASWELL-SSE-NEXT: monitor # sched: [100:0.25]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_monitor:
; HASWELL: # %bb.0:
; HASWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -506,6 +815,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; HASWELL-NEXT: monitor # sched: [100:0.25]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_monitor:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; BROADWELL-SSE-NEXT: movl %esi, %ecx # sched: [1:0.25]
+; BROADWELL-SSE-NEXT: monitor # sched: [100:0.25]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_monitor:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -513,6 +829,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; BROADWELL-NEXT: monitor # sched: [100:0.25]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_monitor:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; SKYLAKE-SSE-NEXT: movl %esi, %ecx # sched: [1:0.25]
+; SKYLAKE-SSE-NEXT: monitor # sched: [100:0.25]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_monitor:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -520,6 +843,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; SKYLAKE-NEXT: monitor # sched: [100:0.25]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_monitor:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; SKX-SSE-NEXT: movl %esi, %ecx # sched: [1:0.25]
+; SKX-SSE-NEXT: monitor # sched: [100:0.25]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_monitor:
; SKX: # %bb.0:
; SKX-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -527,6 +857,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; SKX-NEXT: monitor # sched: [100:0.25]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_monitor:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
+; BTVER2-SSE-NEXT: movl %esi, %ecx # sched: [1:0.50]
+; BTVER2-SSE-NEXT: monitor # sched: [100:0.50]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_monitor:
; BTVER2: # %bb.0:
; BTVER2-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
@@ -534,6 +871,13 @@ define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
; BTVER2-NEXT: monitor # sched: [100:0.50]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_monitor:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: leaq (%rdi), %rax # sched: [1:0.25]
+; ZNVER1-SSE-NEXT: movl %esi, %ecx # sched: [1:0.25]
+; ZNVER1-SSE-NEXT: monitor # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_monitor:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: leaq (%rdi), %rax # sched: [1:0.25]
@@ -569,6 +913,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; SLM-NEXT: movapd %xmm1, %xmm0 # sched: [1:0.50]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_movddup:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
+; SANDY-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [6:0.50]
+; SANDY-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_movddup:
; SANDY: # %bb.0:
; SANDY-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
@@ -576,6 +927,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; SANDY-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_movddup:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
+; HASWELL-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [5:0.50]
+; HASWELL-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_movddup:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
@@ -583,6 +941,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; HASWELL-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_movddup:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
+; BROADWELL-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [5:0.50]
+; BROADWELL-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_movddup:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
@@ -590,6 +955,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; BROADWELL-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_movddup:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
+; SKYLAKE-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [5:0.50]
+; SKYLAKE-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [4:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_movddup:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
@@ -597,6 +969,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; SKYLAKE-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [4:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_movddup:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
+; SKX-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [5:0.50]
+; SKX-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [4:0.33]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_movddup:
; SKX: # %bb.0:
; SKX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
@@ -604,6 +983,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; SKX-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [4:0.33]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_movddup:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:0.50]
+; BTVER2-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [6:1.00]
+; BTVER2-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_movddup:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [6:1.00]
@@ -611,6 +997,13 @@ define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
; BTVER2-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_movddup:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:0.50]
+; ZNVER1-SSE-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [8:0.50]
+; ZNVER1-SSE-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_movddup:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [8:0.50]
@@ -648,6 +1041,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; SLM-NEXT: movaps %xmm1, %xmm0 # sched: [1:0.50]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_movshdup:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
+; SANDY-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
+; SANDY-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_movshdup:
; SANDY: # %bb.0:
; SANDY-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
@@ -655,6 +1055,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_movshdup:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
+; HASWELL-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
+; HASWELL-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_movshdup:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
@@ -662,6 +1069,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_movshdup:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
+; BROADWELL-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [5:0.50]
+; BROADWELL-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_movshdup:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
@@ -669,6 +1083,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_movshdup:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
+; SKYLAKE-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
+; SKYLAKE-SSE-NEXT: addps %xmm1, %xmm0 # sched: [4:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_movshdup:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
@@ -676,6 +1097,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; SKYLAKE-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_movshdup:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
+; SKX-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
+; SKX-SSE-NEXT: addps %xmm1, %xmm0 # sched: [4:0.33]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_movshdup:
; SKX: # %bb.0:
; SKX-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
@@ -683,6 +1111,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; SKX-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_movshdup:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:0.50]
+; BTVER2-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:1.00]
+; BTVER2-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_movshdup:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:1.00]
@@ -690,6 +1125,13 @@ define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
; BTVER2-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_movshdup:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:0.50]
+; ZNVER1-SSE-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [8:0.50]
+; ZNVER1-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_movshdup:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [8:0.50]
@@ -727,6 +1169,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; SLM-NEXT: movaps %xmm1, %xmm0 # sched: [1:0.50]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_movsldup:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
+; SANDY-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
+; SANDY-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_movsldup:
; SANDY: # %bb.0:
; SANDY-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
@@ -734,6 +1183,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_movsldup:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
+; HASWELL-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
+; HASWELL-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_movsldup:
; HASWELL: # %bb.0:
; HASWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
@@ -741,6 +1197,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_movsldup:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
+; BROADWELL-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [5:0.50]
+; BROADWELL-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_movsldup:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
@@ -748,6 +1211,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_movsldup:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
+; SKYLAKE-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
+; SKYLAKE-SSE-NEXT: addps %xmm1, %xmm0 # sched: [4:0.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_movsldup:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
@@ -755,6 +1225,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; SKYLAKE-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_movsldup:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
+; SKX-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
+; SKX-SSE-NEXT: addps %xmm1, %xmm0 # sched: [4:0.33]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_movsldup:
; SKX: # %bb.0:
; SKX-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
@@ -762,6 +1239,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; SKX-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_movsldup:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:0.50]
+; BTVER2-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:1.00]
+; BTVER2-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_movsldup:
; BTVER2: # %bb.0:
; BTVER2-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:1.00]
@@ -769,6 +1253,13 @@ define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
; BTVER2-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_movsldup:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [100:?]
+; ZNVER1-SSE-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [100:?]
+; ZNVER1-SSE-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_movsldup:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [8:0.50]
@@ -804,6 +1295,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; SLM-NEXT: mwait # sched: [100:1.00]
; SLM-NEXT: retq # sched: [4:1.00]
;
+; SANDY-SSE-LABEL: test_mwait:
+; SANDY-SSE: # %bb.0:
+; SANDY-SSE-NEXT: movl %edi, %ecx # sched: [1:0.33]
+; SANDY-SSE-NEXT: movl %esi, %eax # sched: [1:0.33]
+; SANDY-SSE-NEXT: mwait # sched: [100:0.33]
+; SANDY-SSE-NEXT: retq # sched: [1:1.00]
+;
; SANDY-LABEL: test_mwait:
; SANDY: # %bb.0:
; SANDY-NEXT: movl %edi, %ecx # sched: [1:0.33]
@@ -811,6 +1309,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; SANDY-NEXT: mwait # sched: [100:0.33]
; SANDY-NEXT: retq # sched: [1:1.00]
;
+; HASWELL-SSE-LABEL: test_mwait:
+; HASWELL-SSE: # %bb.0:
+; HASWELL-SSE-NEXT: movl %edi, %ecx # sched: [1:0.25]
+; HASWELL-SSE-NEXT: movl %esi, %eax # sched: [1:0.25]
+; HASWELL-SSE-NEXT: mwait # sched: [20:2.50]
+; HASWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; HASWELL-LABEL: test_mwait:
; HASWELL: # %bb.0:
; HASWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
@@ -818,6 +1323,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; HASWELL-NEXT: mwait # sched: [20:2.50]
; HASWELL-NEXT: retq # sched: [7:1.00]
;
+; BROADWELL-SSE-LABEL: test_mwait:
+; BROADWELL-SSE: # %bb.0:
+; BROADWELL-SSE-NEXT: movl %edi, %ecx # sched: [1:0.25]
+; BROADWELL-SSE-NEXT: movl %esi, %eax # sched: [1:0.25]
+; BROADWELL-SSE-NEXT: mwait # sched: [100:0.25]
+; BROADWELL-SSE-NEXT: retq # sched: [7:1.00]
+;
; BROADWELL-LABEL: test_mwait:
; BROADWELL: # %bb.0:
; BROADWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
@@ -825,6 +1337,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; BROADWELL-NEXT: mwait # sched: [100:0.25]
; BROADWELL-NEXT: retq # sched: [7:1.00]
;
+; SKYLAKE-SSE-LABEL: test_mwait:
+; SKYLAKE-SSE: # %bb.0:
+; SKYLAKE-SSE-NEXT: movl %edi, %ecx # sched: [1:0.25]
+; SKYLAKE-SSE-NEXT: movl %esi, %eax # sched: [1:0.25]
+; SKYLAKE-SSE-NEXT: mwait # sched: [20:2.50]
+; SKYLAKE-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKYLAKE-LABEL: test_mwait:
; SKYLAKE: # %bb.0:
; SKYLAKE-NEXT: movl %edi, %ecx # sched: [1:0.25]
@@ -832,6 +1351,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; SKYLAKE-NEXT: mwait # sched: [20:2.50]
; SKYLAKE-NEXT: retq # sched: [7:1.00]
;
+; SKX-SSE-LABEL: test_mwait:
+; SKX-SSE: # %bb.0:
+; SKX-SSE-NEXT: movl %edi, %ecx # sched: [1:0.25]
+; SKX-SSE-NEXT: movl %esi, %eax # sched: [1:0.25]
+; SKX-SSE-NEXT: mwait # sched: [20:2.50]
+; SKX-SSE-NEXT: retq # sched: [7:1.00]
+;
; SKX-LABEL: test_mwait:
; SKX: # %bb.0:
; SKX-NEXT: movl %edi, %ecx # sched: [1:0.25]
@@ -839,6 +1365,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; SKX-NEXT: mwait # sched: [20:2.50]
; SKX-NEXT: retq # sched: [7:1.00]
;
+; BTVER2-SSE-LABEL: test_mwait:
+; BTVER2-SSE: # %bb.0:
+; BTVER2-SSE-NEXT: movl %edi, %ecx # sched: [1:0.50]
+; BTVER2-SSE-NEXT: movl %esi, %eax # sched: [1:0.50]
+; BTVER2-SSE-NEXT: mwait # sched: [100:0.50]
+; BTVER2-SSE-NEXT: retq # sched: [4:1.00]
+;
; BTVER2-LABEL: test_mwait:
; BTVER2: # %bb.0:
; BTVER2-NEXT: movl %edi, %ecx # sched: [1:0.50]
@@ -846,6 +1379,13 @@ define void @test_mwait(i32 %a0, i32 %a1) {
; BTVER2-NEXT: mwait # sched: [100:0.50]
; BTVER2-NEXT: retq # sched: [4:1.00]
;
+; ZNVER1-SSE-LABEL: test_mwait:
+; ZNVER1-SSE: # %bb.0:
+; ZNVER1-SSE-NEXT: movl %edi, %ecx # sched: [1:0.25]
+; ZNVER1-SSE-NEXT: movl %esi, %eax # sched: [1:0.25]
+; ZNVER1-SSE-NEXT: mwait # sched: [100:?]
+; ZNVER1-SSE-NEXT: retq # sched: [1:0.50]
+;
; ZNVER1-LABEL: test_mwait:
; ZNVER1: # %bb.0:
; ZNVER1-NEXT: movl %edi, %ecx # sched: [1:0.25]
OpenPOWER on IntegriCloud