diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/sse2-schedule.ll')
-rw-r--r-- | llvm/test/CodeGen/X86/sse2-schedule.ll | 916 |
1 files changed, 458 insertions, 458 deletions
diff --git a/llvm/test/CodeGen/X86/sse2-schedule.ll b/llvm/test/CodeGen/X86/sse2-schedule.ll index af55e576137..fbc21390d2d 100644 --- a/llvm/test/CodeGen/X86/sse2-schedule.ll +++ b/llvm/test/CodeGen/X86/sse2-schedule.ll @@ -1,5 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC +; 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 @@ -12,9 +12,9 @@ define <2 x double> @test_addpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_addpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: addpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: addpd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_addpd: ; ATOM: # BB#0: @@ -60,9 +60,9 @@ define <2 x double> @test_addpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define double @test_addsd(double %a0, double %a1, double *%a2) { ; GENERIC-LABEL: test_addsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: addsd %xmm1, %xmm0 -; GENERIC-NEXT: addsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: addsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: addsd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_addsd: ; ATOM: # BB#0: @@ -108,10 +108,10 @@ define double @test_addsd(double %a0, double %a1, double *%a2) { define <2 x double> @test_andpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_andpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: andpd %xmm1, %xmm0 -; GENERIC-NEXT: andpd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: andpd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: andpd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_andpd: ; ATOM: # BB#0: @@ -168,10 +168,10 @@ define <2 x double> @test_andpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define <2 x double> @test_andnotpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_andnotpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: andnpd %xmm1, %xmm0 -; GENERIC-NEXT: andnpd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: andnpd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: andnpd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_andnotpd: ; ATOM: # BB#0: @@ -230,10 +230,10 @@ define <2 x double> @test_andnotpd(<2 x double> %a0, <2 x double> %a1, <2 x doub define <2 x double> @test_cmppd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_cmppd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cmpeqpd %xmm0, %xmm1 -; GENERIC-NEXT: cmpeqpd (%rdi), %xmm0 -; GENERIC-NEXT: orpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cmpeqpd %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: cmpeqpd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: orpd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cmppd: ; ATOM: # BB#0: @@ -288,9 +288,9 @@ define <2 x double> @test_cmppd(<2 x double> %a0, <2 x double> %a1, <2 x double> define double @test_cmpsd(double %a0, double %a1, double *%a2) { ; GENERIC-LABEL: test_cmpsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cmpeqsd %xmm1, %xmm0 -; GENERIC-NEXT: cmpeqsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cmpeqsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: cmpeqsd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cmpsd: ; ATOM: # BB#0: @@ -341,17 +341,17 @@ declare <2 x double> @llvm.x86.sse2.cmp.sd(<2 x double>, <2 x double>, i8) nounw define i32 @test_comisd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_comisd: ; GENERIC: # BB#0: -; GENERIC-NEXT: comisd %xmm1, %xmm0 -; GENERIC-NEXT: setnp %al -; GENERIC-NEXT: sete %cl -; GENERIC-NEXT: andb %al, %cl -; GENERIC-NEXT: comisd (%rdi), %xmm0 -; GENERIC-NEXT: setnp %al -; GENERIC-NEXT: sete %dl -; GENERIC-NEXT: andb %al, %dl -; GENERIC-NEXT: orb %cl, %dl -; GENERIC-NEXT: movzbl %dl, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: comisd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: setnp %al # sched: [1:1.00] +; GENERIC-NEXT: sete %cl # sched: [1:1.00] +; GENERIC-NEXT: andb %al, %cl # sched: [1:0.33] +; GENERIC-NEXT: comisd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: setnp %al # sched: [1:1.00] +; GENERIC-NEXT: sete %dl # sched: [1:1.00] +; GENERIC-NEXT: andb %al, %dl # sched: [1:0.33] +; GENERIC-NEXT: orb %cl, %dl # sched: [1:0.33] +; GENERIC-NEXT: movzbl %dl, %eax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_comisd: ; ATOM: # BB#0: @@ -447,10 +447,10 @@ declare i32 @llvm.x86.sse2.comieq.sd(<2 x double>, <2 x double>) nounwind readno define <2 x double> @test_cvtdq2pd(<4 x i32> %a0, <4 x i32> *%a1) { ; GENERIC-LABEL: test_cvtdq2pd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtdq2pd %xmm0, %xmm1 -; GENERIC-NEXT: cvtdq2pd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtdq2pd %xmm0, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvtdq2pd (%rdi), %xmm0 # sched: [10:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtdq2pd: ; ATOM: # BB#0: @@ -505,10 +505,10 @@ define <2 x double> @test_cvtdq2pd(<4 x i32> %a0, <4 x i32> *%a1) { define <4 x float> @test_cvtdq2ps(<4 x i32> %a0, <4 x i32> *%a1) { ; GENERIC-LABEL: test_cvtdq2ps: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtdq2ps %xmm0, %xmm1 -; GENERIC-NEXT: cvtdq2ps (%rdi), %xmm0 -; GENERIC-NEXT: addps %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtdq2ps %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: cvtdq2ps (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtdq2ps: ; ATOM: # BB#0: @@ -562,10 +562,10 @@ define <4 x float> @test_cvtdq2ps(<4 x i32> %a0, <4 x i32> *%a1) { define <4 x i32> @test_cvtpd2dq(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_cvtpd2dq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtpd2dq %xmm0, %xmm1 -; GENERIC-NEXT: cvtpd2dq (%rdi), %xmm0 -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtpd2dq %xmm0, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvtpd2dq (%rdi), %xmm0 # sched: [10:1.00] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtpd2dq: ; ATOM: # BB#0: @@ -620,10 +620,10 @@ declare <4 x i32> @llvm.x86.sse2.cvtpd2dq(<2 x double>) nounwind readnone define <4 x float> @test_cvtpd2ps(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_cvtpd2ps: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtpd2ps %xmm0, %xmm1 -; GENERIC-NEXT: cvtpd2ps (%rdi), %xmm0 -; GENERIC-NEXT: addps %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtpd2ps %xmm0, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvtpd2ps (%rdi), %xmm0 # sched: [10:1.00] +; GENERIC-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtpd2ps: ; ATOM: # BB#0: @@ -678,10 +678,10 @@ declare <4 x float> @llvm.x86.sse2.cvtpd2ps(<2 x double>) nounwind readnone define <4 x i32> @test_cvtps2dq(<4 x float> %a0, <4 x float> *%a1) { ; GENERIC-LABEL: test_cvtps2dq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtps2dq %xmm0, %xmm1 -; GENERIC-NEXT: cvtps2dq (%rdi), %xmm0 -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtps2dq %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: cvtps2dq (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtps2dq: ; ATOM: # BB#0: @@ -736,10 +736,10 @@ declare <4 x i32> @llvm.x86.sse2.cvtps2dq(<4 x float>) nounwind readnone define <2 x double> @test_cvtps2pd(<4 x float> %a0, <4 x float> *%a1) { ; GENERIC-LABEL: test_cvtps2pd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtps2pd %xmm0, %xmm1 -; GENERIC-NEXT: cvtps2pd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtps2pd %xmm0, %xmm1 # sched: [2:1.00] +; GENERIC-NEXT: cvtps2pd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtps2pd: ; ATOM: # BB#0: @@ -795,10 +795,10 @@ define <2 x double> @test_cvtps2pd(<4 x float> %a0, <4 x float> *%a1) { define i32 @test_cvtsd2si(double %a0, double *%a1) { ; GENERIC-LABEL: test_cvtsd2si: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtsd2si %xmm0, %ecx -; GENERIC-NEXT: cvtsd2si (%rdi), %eax -; GENERIC-NEXT: addl %ecx, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtsd2si %xmm0, %ecx # sched: [5:1.00] +; GENERIC-NEXT: cvtsd2si (%rdi), %eax # sched: [9:1.00] +; GENERIC-NEXT: addl %ecx, %eax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtsd2si: ; ATOM: # BB#0: @@ -854,10 +854,10 @@ declare i32 @llvm.x86.sse2.cvtsd2si(<2 x double>) nounwind readnone define i64 @test_cvtsd2siq(double %a0, double *%a1) { ; GENERIC-LABEL: test_cvtsd2siq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtsd2si %xmm0, %rcx -; GENERIC-NEXT: cvtsd2si (%rdi), %rax -; GENERIC-NEXT: addq %rcx, %rax -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtsd2si %xmm0, %rcx # sched: [5:1.00] +; GENERIC-NEXT: cvtsd2si (%rdi), %rax # sched: [9:1.00] +; GENERIC-NEXT: addq %rcx, %rax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtsd2siq: ; ATOM: # BB#0: @@ -913,11 +913,11 @@ declare i64 @llvm.x86.sse2.cvtsd2si64(<2 x double>) nounwind readnone define float @test_cvtsd2ss(double %a0, double *%a1) { ; GENERIC-LABEL: test_cvtsd2ss: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtsd2ss %xmm0, %xmm1 -; GENERIC-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero -; GENERIC-NEXT: cvtsd2ss %xmm0, %xmm0 -; GENERIC-NEXT: addss %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtsd2ss %xmm0, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero sched: [4:0.50] +; GENERIC-NEXT: cvtsd2ss %xmm0, %xmm0 # sched: [4:1.00] +; GENERIC-NEXT: addss %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtsd2ss: ; ATOM: # BB#0: @@ -977,10 +977,10 @@ define float @test_cvtsd2ss(double %a0, double *%a1) { define double @test_cvtsi2sd(i32 %a0, i32 *%a1) { ; GENERIC-LABEL: test_cvtsi2sd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtsi2sdl %edi, %xmm1 -; GENERIC-NEXT: cvtsi2sdl (%rsi), %xmm0 -; GENERIC-NEXT: addsd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtsi2sdl %edi, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvtsi2sdl (%rsi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: addsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtsi2sd: ; ATOM: # BB#0: @@ -1033,10 +1033,10 @@ define double @test_cvtsi2sd(i32 %a0, i32 *%a1) { define double @test_cvtsi2sdq(i64 %a0, i64 *%a1) { ; GENERIC-LABEL: test_cvtsi2sdq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtsi2sdq %rdi, %xmm1 -; GENERIC-NEXT: cvtsi2sdq (%rsi), %xmm0 -; GENERIC-NEXT: addsd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtsi2sdq %rdi, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvtsi2sdq (%rsi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: addsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtsi2sdq: ; ATOM: # BB#0: @@ -1091,11 +1091,11 @@ define double @test_cvtsi2sdq(i64 %a0, i64 *%a1) { define double @test_cvtss2sd(float %a0, float *%a1) { ; GENERIC-LABEL: test_cvtss2sd: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvtss2sd %xmm0, %xmm1 -; GENERIC-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; GENERIC-NEXT: cvtss2sd %xmm0, %xmm0 -; GENERIC-NEXT: addsd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvtss2sd %xmm0, %xmm1 # sched: [1:1.00] +; GENERIC-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero sched: [6:0.50] +; GENERIC-NEXT: cvtss2sd %xmm0, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: addsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvtss2sd: ; ATOM: # BB#0: @@ -1155,10 +1155,10 @@ define double @test_cvtss2sd(float %a0, float *%a1) { define <4 x i32> @test_cvttpd2dq(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_cvttpd2dq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvttpd2dq %xmm0, %xmm1 -; GENERIC-NEXT: cvttpd2dq (%rdi), %xmm0 -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvttpd2dq %xmm0, %xmm1 # sched: [4:1.00] +; GENERIC-NEXT: cvttpd2dq (%rdi), %xmm0 # sched: [10:1.00] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvttpd2dq: ; ATOM: # BB#0: @@ -1214,10 +1214,10 @@ define <4 x i32> @test_cvttpd2dq(<2 x double> %a0, <2 x double> *%a1) { define <4 x i32> @test_cvttps2dq(<4 x float> %a0, <4 x float> *%a1) { ; GENERIC-LABEL: test_cvttps2dq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvttps2dq %xmm0, %xmm1 -; GENERIC-NEXT: cvttps2dq (%rdi), %xmm0 -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvttps2dq %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: cvttps2dq (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvttps2dq: ; ATOM: # BB#0: @@ -1271,10 +1271,10 @@ define <4 x i32> @test_cvttps2dq(<4 x float> %a0, <4 x float> *%a1) { define i32 @test_cvttsd2si(double %a0, double *%a1) { ; GENERIC-LABEL: test_cvttsd2si: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvttsd2si %xmm0, %ecx -; GENERIC-NEXT: cvttsd2si (%rdi), %eax -; GENERIC-NEXT: addl %ecx, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvttsd2si %xmm0, %ecx # sched: [5:1.00] +; GENERIC-NEXT: cvttsd2si (%rdi), %eax # sched: [9:1.00] +; GENERIC-NEXT: addl %ecx, %eax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvttsd2si: ; ATOM: # BB#0: @@ -1327,10 +1327,10 @@ define i32 @test_cvttsd2si(double %a0, double *%a1) { define i64 @test_cvttsd2siq(double %a0, double *%a1) { ; GENERIC-LABEL: test_cvttsd2siq: ; GENERIC: # BB#0: -; GENERIC-NEXT: cvttsd2si %xmm0, %rcx -; GENERIC-NEXT: cvttsd2si (%rdi), %rax -; GENERIC-NEXT: addq %rcx, %rax -; GENERIC-NEXT: retq +; GENERIC-NEXT: cvttsd2si %xmm0, %rcx # sched: [5:1.00] +; GENERIC-NEXT: cvttsd2si (%rdi), %rax # sched: [9:1.00] +; GENERIC-NEXT: addq %rcx, %rax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_cvttsd2siq: ; ATOM: # BB#0: @@ -1383,9 +1383,9 @@ define i64 @test_cvttsd2siq(double %a0, double *%a1) { define <2 x double> @test_divpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_divpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: divpd %xmm1, %xmm0 -; GENERIC-NEXT: divpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: divpd %xmm1, %xmm0 # sched: [22:1.00] +; GENERIC-NEXT: divpd (%rdi), %xmm0 # sched: [28:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_divpd: ; ATOM: # BB#0: @@ -1431,9 +1431,9 @@ define <2 x double> @test_divpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define double @test_divsd(double %a0, double %a1, double *%a2) { ; GENERIC-LABEL: test_divsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: divsd %xmm1, %xmm0 -; GENERIC-NEXT: divsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: divsd %xmm1, %xmm0 # sched: [22:1.00] +; GENERIC-NEXT: divsd (%rdi), %xmm0 # sched: [28:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_divsd: ; ATOM: # BB#0: @@ -1479,8 +1479,8 @@ define double @test_divsd(double %a0, double %a1, double *%a2) { define void @test_lfence() { ; GENERIC-LABEL: test_lfence: ; GENERIC: # BB#0: -; GENERIC-NEXT: lfence -; GENERIC-NEXT: retq +; GENERIC-NEXT: lfence # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_lfence: ; ATOM: # BB#0: @@ -1525,8 +1525,8 @@ declare void @llvm.x86.sse2.lfence() nounwind readnone define void @test_mfence() { ; GENERIC-LABEL: test_mfence: ; GENERIC: # BB#0: -; GENERIC-NEXT: mfence -; GENERIC-NEXT: retq +; GENERIC-NEXT: mfence # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_mfence: ; ATOM: # BB#0: @@ -1571,8 +1571,8 @@ declare void @llvm.x86.sse2.mfence() nounwind readnone define void @test_maskmovdqu(<16 x i8> %a0, <16 x i8> %a1, i8* %a2) { ; GENERIC-LABEL: test_maskmovdqu: ; GENERIC: # BB#0: -; GENERIC-NEXT: maskmovdqu %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: maskmovdqu %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_maskmovdqu: ; ATOM: # BB#0: @@ -1615,9 +1615,9 @@ declare void @llvm.x86.sse2.maskmov.dqu(<16 x i8>, <16 x i8>, i8*) nounwind define <2 x double> @test_maxpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_maxpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: maxpd %xmm1, %xmm0 -; GENERIC-NEXT: maxpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: maxpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: maxpd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_maxpd: ; ATOM: # BB#0: @@ -1664,9 +1664,9 @@ declare <2 x double> @llvm.x86.sse2.max.pd(<2 x double>, <2 x double>) nounwind define <2 x double> @test_maxsd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_maxsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: maxsd %xmm1, %xmm0 -; GENERIC-NEXT: maxsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: maxsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: maxsd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_maxsd: ; ATOM: # BB#0: @@ -1713,9 +1713,9 @@ declare <2 x double> @llvm.x86.sse2.max.sd(<2 x double>, <2 x double>) nounwind define <2 x double> @test_minpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_minpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: minpd %xmm1, %xmm0 -; GENERIC-NEXT: minpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: minpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: minpd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_minpd: ; ATOM: # BB#0: @@ -1762,9 +1762,9 @@ declare <2 x double> @llvm.x86.sse2.min.pd(<2 x double>, <2 x double>) nounwind define <2 x double> @test_minsd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_minsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: minsd %xmm1, %xmm0 -; GENERIC-NEXT: minsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: minsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: minsd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_minsd: ; ATOM: # BB#0: @@ -1811,10 +1811,10 @@ declare <2 x double> @llvm.x86.sse2.min.sd(<2 x double>, <2 x double>) nounwind define void @test_movapd(<2 x double> *%a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_movapd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movapd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm0, %xmm0 -; GENERIC-NEXT: movapd %xmm0, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movapd (%rdi), %xmm0 # sched: [6:0.50] +; GENERIC-NEXT: addpd %xmm0, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: movapd %xmm0, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movapd: ; ATOM: # BB#0: @@ -1866,10 +1866,10 @@ define void @test_movapd(<2 x double> *%a0, <2 x double> *%a1) { define void @test_movdqa(<2 x i64> *%a0, <2 x i64> *%a1) { ; GENERIC-LABEL: test_movdqa: ; GENERIC: # BB#0: -; GENERIC-NEXT: movdqa (%rdi), %xmm0 -; GENERIC-NEXT: paddq %xmm0, %xmm0 -; GENERIC-NEXT: movdqa %xmm0, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movdqa (%rdi), %xmm0 # sched: [6:0.50] +; GENERIC-NEXT: paddq %xmm0, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: movdqa %xmm0, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movdqa: ; ATOM: # BB#0: @@ -1921,10 +1921,10 @@ define void @test_movdqa(<2 x i64> *%a0, <2 x i64> *%a1) { define void @test_movdqu(<2 x i64> *%a0, <2 x i64> *%a1) { ; GENERIC-LABEL: test_movdqu: ; GENERIC: # BB#0: -; GENERIC-NEXT: movdqu (%rdi), %xmm0 -; GENERIC-NEXT: paddq %xmm0, %xmm0 -; GENERIC-NEXT: movdqu %xmm0, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movdqu (%rdi), %xmm0 # sched: [6:0.50] +; GENERIC-NEXT: paddq %xmm0, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: movdqu %xmm0, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movdqu: ; ATOM: # BB#0: @@ -1976,13 +1976,13 @@ define void @test_movdqu(<2 x i64> *%a0, <2 x i64> *%a1) { define i32 @test_movd(<4 x i32> %a0, i32 %a1, i32 *%a2) { ; GENERIC-LABEL: test_movd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movd %edi, %xmm1 -; GENERIC-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero -; GENERIC-NEXT: paddd %xmm0, %xmm1 -; GENERIC-NEXT: paddd %xmm0, %xmm2 -; GENERIC-NEXT: movd %xmm2, %eax -; GENERIC-NEXT: movd %xmm1, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movd %edi, %xmm1 # sched: [1:1.00] +; GENERIC-NEXT: movd {{.*#+}} xmm2 = mem[0],zero,zero,zero sched: [6:0.50] +; GENERIC-NEXT: paddd %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: paddd %xmm0, %xmm2 # sched: [1:0.50] +; GENERIC-NEXT: movd %xmm2, %eax # sched: [2:1.00] +; GENERIC-NEXT: movd %xmm1, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movd: ; ATOM: # BB#0: @@ -2057,13 +2057,13 @@ define i32 @test_movd(<4 x i32> %a0, i32 %a1, i32 *%a2) { define i64 @test_movd_64(<2 x i64> %a0, i64 %a1, i64 *%a2) { ; GENERIC-LABEL: test_movd_64: ; GENERIC: # BB#0: -; GENERIC-NEXT: movq %rdi, %xmm1 -; GENERIC-NEXT: movq {{.*#+}} xmm2 = mem[0],zero -; GENERIC-NEXT: paddq %xmm0, %xmm1 -; GENERIC-NEXT: paddq %xmm0, %xmm2 -; GENERIC-NEXT: movq %xmm2, %rax -; GENERIC-NEXT: movq %xmm1, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movq %rdi, %xmm1 # sched: [1:1.00] +; GENERIC-NEXT: movq {{.*#+}} xmm2 = mem[0],zero sched: [4:0.50] +; GENERIC-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: paddq %xmm0, %xmm2 # sched: [1:0.50] +; GENERIC-NEXT: movq %xmm2, %rax # sched: [2:1.00] +; GENERIC-NEXT: movq %xmm1, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movd_64: ; ATOM: # BB#0: @@ -2138,10 +2138,10 @@ define i64 @test_movd_64(<2 x i64> %a0, i64 %a1, i64 *%a2) { define void @test_movhpd(<2 x double> %a0, <2 x double> %a1, x86_mmx *%a2) { ; GENERIC-LABEL: test_movhpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] -; GENERIC-NEXT: addpd %xmm0, %xmm1 -; GENERIC-NEXT: movhpd %xmm1, (%rdi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movhpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: movhpd %xmm1, (%rdi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movhpd: ; ATOM: # BB#0: @@ -2196,10 +2196,10 @@ define void @test_movhpd(<2 x double> %a0, <2 x double> %a1, x86_mmx *%a2) { define void @test_movlpd(<2 x double> %a0, <2 x double> %a1, x86_mmx *%a2) { ; GENERIC-LABEL: test_movlpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] -; GENERIC-NEXT: addpd %xmm0, %xmm1 -; GENERIC-NEXT: movlpd %xmm1, (%rdi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movlpd {{.*#+}} xmm1 = mem[0],xmm1[1] sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: movlpd %xmm1, (%rdi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movlpd: ; ATOM: # BB#0: @@ -2254,8 +2254,8 @@ define void @test_movlpd(<2 x double> %a0, <2 x double> %a1, x86_mmx *%a2) { define i32 @test_movmskpd(<2 x double> %a0) { ; GENERIC-LABEL: test_movmskpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movmskpd %xmm0, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: movmskpd %xmm0, %eax # sched: [2:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movmskpd: ; ATOM: # BB#0: @@ -2296,9 +2296,9 @@ declare i32 @llvm.x86.sse2.movmsk.pd(<2 x double>) nounwind readnone define void @test_movntdqa(<2 x i64> %a0, <2 x i64> *%a1) { ; GENERIC-LABEL: test_movntdqa: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddq %xmm0, %xmm0 -; GENERIC-NEXT: movntdq %xmm0, (%rdi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddq %xmm0, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: movntdq %xmm0, (%rdi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movntdqa: ; ATOM: # BB#0: @@ -2345,9 +2345,9 @@ define void @test_movntdqa(<2 x i64> %a0, <2 x i64> *%a1) { define void @test_movntpd(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_movntpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: addpd %xmm0, %xmm0 -; GENERIC-NEXT: movntpd %xmm0, (%rdi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: addpd %xmm0, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: movntpd %xmm0, (%rdi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movntpd: ; ATOM: # BB#0: @@ -2392,10 +2392,10 @@ define void @test_movntpd(<2 x double> %a0, <2 x double> *%a1) { define <2 x i64> @test_movq_mem(<2 x i64> %a0, i64 *%a1) { ; GENERIC-LABEL: test_movq_mem: ; GENERIC: # BB#0: -; GENERIC-NEXT: movq {{.*#+}} xmm1 = mem[0],zero -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: movq %xmm0, (%rdi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movq {{.*#+}} xmm1 = mem[0],zero sched: [4:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: movq %xmm0, (%rdi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movq_mem: ; ATOM: # BB#0: @@ -2449,9 +2449,9 @@ define <2 x i64> @test_movq_mem(<2 x i64> %a0, i64 *%a1) { define <2 x i64> @test_movq_reg(<2 x i64> %a0, <2 x i64> %a1) { ; GENERIC-LABEL: test_movq_reg: ; GENERIC: # BB#0: -; GENERIC-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: movq {{.*#+}} xmm0 = xmm0[0],zero sched: [1:1.00] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movq_reg: ; ATOM: # BB#0: @@ -2498,10 +2498,10 @@ define <2 x i64> @test_movq_reg(<2 x i64> %a0, <2 x i64> %a1) { define void @test_movsd_mem(double* %a0, double* %a1) { ; GENERIC-LABEL: test_movsd_mem: ; GENERIC: # BB#0: -; GENERIC-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero -; GENERIC-NEXT: addsd %xmm0, %xmm0 -; GENERIC-NEXT: movsd %xmm0, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero sched: [4:0.50] +; GENERIC-NEXT: addsd %xmm0, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: movsd %xmm0, (%rsi) # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movsd_mem: ; ATOM: # BB#0: @@ -2553,9 +2553,9 @@ define void @test_movsd_mem(double* %a0, double* %a1) { define <2 x double> @test_movsd_reg(<2 x double> %a0, <2 x double> %a1) { ; GENERIC-LABEL: test_movsd_reg: ; GENERIC: # BB#0: -; GENERIC-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0] -; GENERIC-NEXT: movapd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0] sched: [1:1.00] +; GENERIC-NEXT: movapd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movsd_reg: ; ATOM: # BB#0: @@ -2599,10 +2599,10 @@ define <2 x double> @test_movsd_reg(<2 x double> %a0, <2 x double> %a1) { define void @test_movupd(<2 x double> *%a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_movupd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movupd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm0, %xmm0 -; GENERIC-NEXT: movupd %xmm0, (%rsi) -; GENERIC-NEXT: retq +; GENERIC-NEXT: movupd (%rdi), %xmm0 # sched: [6:0.50] +; GENERIC-NEXT: addpd %xmm0, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: movupd %xmm0, (%rsi) # sched: [5:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_movupd: ; ATOM: # BB#0: @@ -2654,9 +2654,9 @@ define void @test_movupd(<2 x double> *%a0, <2 x double> *%a1) { define <2 x double> @test_mulpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_mulpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: mulpd %xmm1, %xmm0 -; GENERIC-NEXT: mulpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: mulpd %xmm1, %xmm0 # sched: [5:1.00] +; GENERIC-NEXT: mulpd (%rdi), %xmm0 # sched: [11:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_mulpd: ; ATOM: # BB#0: @@ -2702,9 +2702,9 @@ define <2 x double> @test_mulpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define double @test_mulsd(double %a0, double %a1, double *%a2) { ; GENERIC-LABEL: test_mulsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: mulsd %xmm1, %xmm0 -; GENERIC-NEXT: mulsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: mulsd %xmm1, %xmm0 # sched: [5:1.00] +; GENERIC-NEXT: mulsd (%rdi), %xmm0 # sched: [11:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_mulsd: ; ATOM: # BB#0: @@ -2750,10 +2750,10 @@ define double @test_mulsd(double %a0, double %a1, double *%a2) { define <2 x double> @test_orpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_orpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: orpd %xmm1, %xmm0 -; GENERIC-NEXT: orpd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: orpd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: orpd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_orpd: ; ATOM: # BB#0: @@ -2810,9 +2810,9 @@ define <2 x double> @test_orpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define <8 x i16> @test_packssdw(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_packssdw: ; GENERIC: # BB#0: -; GENERIC-NEXT: packssdw %xmm1, %xmm0 -; GENERIC-NEXT: packssdw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: packssdw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: packssdw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_packssdw: ; ATOM: # BB#0: @@ -2868,9 +2868,9 @@ declare <8 x i16> @llvm.x86.sse2.packssdw.128(<4 x i32>, <4 x i32>) nounwind rea define <16 x i8> @test_packsswb(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_packsswb: ; GENERIC: # BB#0: -; GENERIC-NEXT: packsswb %xmm1, %xmm0 -; GENERIC-NEXT: packsswb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: packsswb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: packsswb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_packsswb: ; ATOM: # BB#0: @@ -2926,9 +2926,9 @@ declare <16 x i8> @llvm.x86.sse2.packsswb.128(<8 x i16>, <8 x i16>) nounwind rea define <16 x i8> @test_packuswb(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_packuswb: ; GENERIC: # BB#0: -; GENERIC-NEXT: packuswb %xmm1, %xmm0 -; GENERIC-NEXT: packuswb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: packuswb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: packuswb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_packuswb: ; ATOM: # BB#0: @@ -2984,9 +2984,9 @@ declare <16 x i8> @llvm.x86.sse2.packuswb.128(<8 x i16>, <8 x i16>) nounwind rea define <16 x i8> @test_paddb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_paddb: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddb %xmm1, %xmm0 -; GENERIC-NEXT: paddb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddb: ; ATOM: # BB#0: @@ -3036,9 +3036,9 @@ define <16 x i8> @test_paddb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_paddd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_paddd: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: paddd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddd (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddd: ; ATOM: # BB#0: @@ -3088,9 +3088,9 @@ define <4 x i32> @test_paddd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <2 x i64> @test_paddq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_paddq: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: paddq (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddq (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddq: ; ATOM: # BB#0: @@ -3136,9 +3136,9 @@ define <2 x i64> @test_paddq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <16 x i8> @test_paddsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_paddsb: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddsb %xmm1, %xmm0 -; GENERIC-NEXT: paddsb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddsb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddsb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddsb: ; ATOM: # BB#0: @@ -3189,9 +3189,9 @@ declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone define <8 x i16> @test_paddsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_paddsw: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddsw %xmm1, %xmm0 -; GENERIC-NEXT: paddsw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddsw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddsw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddsw: ; ATOM: # BB#0: @@ -3242,9 +3242,9 @@ declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone define <16 x i8> @test_paddusb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_paddusb: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddusb %xmm1, %xmm0 -; GENERIC-NEXT: paddusb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddusb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddusb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddusb: ; ATOM: # BB#0: @@ -3295,9 +3295,9 @@ declare <16 x i8> @llvm.x86.sse2.paddus.b(<16 x i8>, <16 x i8>) nounwind readnon define <8 x i16> @test_paddusw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_paddusw: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddusw %xmm1, %xmm0 -; GENERIC-NEXT: paddusw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddusw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddusw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddusw: ; ATOM: # BB#0: @@ -3348,9 +3348,9 @@ declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnon define <8 x i16> @test_paddw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_paddw: ; GENERIC: # BB#0: -; GENERIC-NEXT: paddw %xmm1, %xmm0 -; GENERIC-NEXT: paddw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: paddw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: paddw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_paddw: ; ATOM: # BB#0: @@ -3400,10 +3400,10 @@ define <8 x i16> @test_paddw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <2 x i64> @test_pand(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_pand: ; GENERIC: # BB#0: -; GENERIC-NEXT: pand %xmm1, %xmm0 -; GENERIC-NEXT: pand (%rdi), %xmm0 -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pand %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: pand (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pand: ; ATOM: # BB#0: @@ -3456,12 +3456,12 @@ define <2 x i64> @test_pand(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <2 x i64> @test_pandn(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_pandn: ; GENERIC: # BB#0: -; GENERIC-NEXT: pandn %xmm1, %xmm0 -; GENERIC-NEXT: movdqa %xmm0, %xmm1 -; GENERIC-NEXT: pandn (%rdi), %xmm1 -; GENERIC-NEXT: paddq %xmm0, %xmm1 -; GENERIC-NEXT: movdqa %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pandn %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: movdqa %xmm0, %xmm1 # sched: [1:0.33] +; GENERIC-NEXT: pandn (%rdi), %xmm1 # sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: movdqa %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pandn: ; ATOM: # BB#0: @@ -3520,9 +3520,9 @@ define <2 x i64> @test_pandn(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <16 x i8> @test_pavgb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_pavgb: ; GENERIC: # BB#0: -; GENERIC-NEXT: pavgb %xmm1, %xmm0 -; GENERIC-NEXT: pavgb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pavgb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pavgb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pavgb: ; ATOM: # BB#0: @@ -3573,9 +3573,9 @@ declare <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8> %arg0, <16 x i8> %arg1) nounwi define <8 x i16> @test_pavgw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pavgw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pavgw %xmm1, %xmm0 -; GENERIC-NEXT: pavgw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pavgw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pavgw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pavgw: ; ATOM: # BB#0: @@ -3626,10 +3626,10 @@ declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone define <16 x i8> @test_pcmpeqb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_pcmpeqb: ; GENERIC: # BB#0: -; GENERIC-NEXT: pcmpeqb %xmm0, %xmm1 -; GENERIC-NEXT: pcmpeqb (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pcmpeqb %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: pcmpeqb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpeqb: ; ATOM: # BB#0: @@ -3685,10 +3685,10 @@ define <16 x i8> @test_pcmpeqb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_pcmpeqd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_pcmpeqd: ; GENERIC: # BB#0: -; GENERIC-NEXT: pcmpeqd %xmm0, %xmm1 -; GENERIC-NEXT: pcmpeqd (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pcmpeqd %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: pcmpeqd (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpeqd: ; ATOM: # BB#0: @@ -3744,10 +3744,10 @@ define <4 x i32> @test_pcmpeqd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <8 x i16> @test_pcmpeqw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pcmpeqw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pcmpeqw %xmm0, %xmm1 -; GENERIC-NEXT: pcmpeqw (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pcmpeqw %xmm0, %xmm1 # sched: [1:0.50] +; GENERIC-NEXT: pcmpeqw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpeqw: ; ATOM: # BB#0: @@ -3803,11 +3803,11 @@ define <8 x i16> @test_pcmpeqw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <16 x i8> @test_pcmpgtb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_pcmpgtb: ; GENERIC: # BB#0: -; GENERIC-NEXT: movdqa %xmm0, %xmm2 -; GENERIC-NEXT: pcmpgtb %xmm1, %xmm2 -; GENERIC-NEXT: pcmpgtb (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: movdqa %xmm0, %xmm2 # sched: [1:0.33] +; GENERIC-NEXT: pcmpgtb %xmm1, %xmm2 # sched: [1:0.50] +; GENERIC-NEXT: pcmpgtb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm2, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpgtb: ; ATOM: # BB#0: @@ -3863,11 +3863,11 @@ define <16 x i8> @test_pcmpgtb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_pcmpgtd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_pcmpgtd: ; GENERIC: # BB#0: -; GENERIC-NEXT: movdqa %xmm0, %xmm2 -; GENERIC-NEXT: pcmpgtd %xmm1, %xmm2 -; GENERIC-NEXT: pcmpeqd (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: movdqa %xmm0, %xmm2 # sched: [1:0.33] +; GENERIC-NEXT: pcmpgtd %xmm1, %xmm2 # sched: [1:0.50] +; GENERIC-NEXT: pcmpeqd (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm2, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpgtd: ; ATOM: # BB#0: @@ -3923,11 +3923,11 @@ define <4 x i32> @test_pcmpgtd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <8 x i16> @test_pcmpgtw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pcmpgtw: ; GENERIC: # BB#0: -; GENERIC-NEXT: movdqa %xmm0, %xmm2 -; GENERIC-NEXT: pcmpgtw %xmm1, %xmm2 -; GENERIC-NEXT: pcmpgtw (%rdi), %xmm0 -; GENERIC-NEXT: por %xmm2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: movdqa %xmm0, %xmm2 # sched: [1:0.33] +; GENERIC-NEXT: pcmpgtw %xmm1, %xmm2 # sched: [1:0.50] +; GENERIC-NEXT: pcmpgtw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: por %xmm2, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pcmpgtw: ; ATOM: # BB#0: @@ -3983,9 +3983,9 @@ define <8 x i16> @test_pcmpgtw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define i16 @test_pextrw(<8 x i16> %a0) { ; GENERIC-LABEL: test_pextrw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pextrw $6, %xmm0, %eax +; GENERIC-NEXT: pextrw $6, %xmm0, %eax # sched: [3:1.00] ; GENERIC-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill> -; GENERIC-NEXT: retq +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pextrw: ; ATOM: # BB#0: @@ -4029,9 +4029,9 @@ define i16 @test_pextrw(<8 x i16> %a0) { define <8 x i16> @test_pinsrw(<8 x i16> %a0, i16 %a1, i16 *%a2) { ; GENERIC-LABEL: test_pinsrw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pinsrw $1, %edi, %xmm0 -; GENERIC-NEXT: pinsrw $3, (%rsi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pinsrw $1, %edi, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: pinsrw $3, (%rsi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pinsrw: ; ATOM: # BB#0: @@ -4081,9 +4081,9 @@ define <8 x i16> @test_pinsrw(<8 x i16> %a0, i16 %a1, i16 *%a2) { define <4 x i32> @test_pmaddwd(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pmaddwd: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmaddwd %xmm1, %xmm0 -; GENERIC-NEXT: pmaddwd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmaddwd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: pmaddwd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmaddwd: ; ATOM: # BB#0: @@ -4139,9 +4139,9 @@ declare <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16>, <8 x i16>) nounwind readnon define <8 x i16> @test_pmaxsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pmaxsw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmaxsw %xmm1, %xmm0 -; GENERIC-NEXT: pmaxsw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmaxsw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pmaxsw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmaxsw: ; ATOM: # BB#0: @@ -4192,9 +4192,9 @@ declare <8 x i16> @llvm.x86.sse2.pmaxs.w(<8 x i16>, <8 x i16>) nounwind readnone define <16 x i8> @test_pmaxub(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_pmaxub: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmaxub %xmm1, %xmm0 -; GENERIC-NEXT: pmaxub (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmaxub %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pmaxub (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmaxub: ; ATOM: # BB#0: @@ -4245,9 +4245,9 @@ declare <16 x i8> @llvm.x86.sse2.pmaxu.b(<16 x i8>, <16 x i8>) nounwind readnone define <8 x i16> @test_pminsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pminsw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pminsw %xmm1, %xmm0 -; GENERIC-NEXT: pminsw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pminsw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pminsw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pminsw: ; ATOM: # BB#0: @@ -4298,9 +4298,9 @@ declare <8 x i16> @llvm.x86.sse2.pmins.w(<8 x i16>, <8 x i16>) nounwind readnone define <16 x i8> @test_pminub(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_pminub: ; GENERIC: # BB#0: -; GENERIC-NEXT: pminub %xmm1, %xmm0 -; GENERIC-NEXT: pminub (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pminub %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: pminub (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pminub: ; ATOM: # BB#0: @@ -4351,8 +4351,8 @@ declare <16 x i8> @llvm.x86.sse2.pminu.b(<16 x i8>, <16 x i8>) nounwind readnone define i32 @test_pmovmskb(<16 x i8> %a0) { ; GENERIC-LABEL: test_pmovmskb: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmovmskb %xmm0, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmovmskb %xmm0, %eax # sched: [2:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmovmskb: ; ATOM: # BB#0: @@ -4393,9 +4393,9 @@ declare i32 @llvm.x86.sse2.pmovmskb.128(<16 x i8>) nounwind readnone define <8 x i16> @test_pmulhuw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pmulhuw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmulhuw %xmm1, %xmm0 -; GENERIC-NEXT: pmulhuw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmulhuw %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: pmulhuw (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmulhuw: ; ATOM: # BB#0: @@ -4442,9 +4442,9 @@ declare <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16>, <8 x i16>) nounwind readnon define <8 x i16> @test_pmulhw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pmulhw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmulhw %xmm1, %xmm0 -; GENERIC-NEXT: pmulhw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmulhw %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: pmulhw (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmulhw: ; ATOM: # BB#0: @@ -4491,9 +4491,9 @@ declare <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16>, <8 x i16>) nounwind readnone define <8 x i16> @test_pmullw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_pmullw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmullw %xmm1, %xmm0 -; GENERIC-NEXT: pmullw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmullw %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: pmullw (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmullw: ; ATOM: # BB#0: @@ -4539,9 +4539,9 @@ define <8 x i16> @test_pmullw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <2 x i64> @test_pmuludq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_pmuludq: ; GENERIC: # BB#0: -; GENERIC-NEXT: pmuludq %xmm1, %xmm0 -; GENERIC-NEXT: pmuludq (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pmuludq %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: pmuludq (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pmuludq: ; ATOM: # BB#0: @@ -4597,10 +4597,10 @@ declare <2 x i64> @llvm.x86.sse2.pmulu.dq(<4 x i32>, <4 x i32>) nounwind readnon define <2 x i64> @test_por(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_por: ; GENERIC: # BB#0: -; GENERIC-NEXT: por %xmm1, %xmm0 -; GENERIC-NEXT: por (%rdi), %xmm0 -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: por %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: por (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_por: ; ATOM: # BB#0: @@ -4653,9 +4653,9 @@ define <2 x i64> @test_por(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <2 x i64> @test_psadbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_psadbw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psadbw %xmm1, %xmm0 -; GENERIC-NEXT: psadbw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psadbw %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: psadbw (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psadbw: ; ATOM: # BB#0: @@ -4711,10 +4711,10 @@ declare <2 x i64> @llvm.x86.sse2.psad.bw(<16 x i8>, <16 x i8>) nounwind readnone define <4 x i32> @test_pshufd(<4 x i32> %a0, <4 x i32> *%a1) { ; GENERIC-LABEL: test_pshufd: ; GENERIC: # BB#0: -; GENERIC-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2] -; GENERIC-NEXT: pshufd {{.*#+}} xmm0 = mem[3,2,1,0] -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,0,3,2] sched: [1:0.50] +; GENERIC-NEXT: pshufd {{.*#+}} xmm0 = mem[3,2,1,0] sched: [7:0.50] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pshufd: ; ATOM: # BB#0: @@ -4769,10 +4769,10 @@ define <4 x i32> @test_pshufd(<4 x i32> %a0, <4 x i32> *%a1) { define <8 x i16> @test_pshufhw(<8 x i16> %a0, <8 x i16> *%a1) { ; GENERIC-LABEL: test_pshufhw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,5,4,7,6] -; GENERIC-NEXT: pshufhw {{.*#+}} xmm0 = mem[0,1,2,3,7,6,5,4] -; GENERIC-NEXT: paddw %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,5,4,7,6] sched: [1:0.50] +; GENERIC-NEXT: pshufhw {{.*#+}} xmm0 = mem[0,1,2,3,7,6,5,4] sched: [7:0.50] +; GENERIC-NEXT: paddw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pshufhw: ; ATOM: # BB#0: @@ -4827,10 +4827,10 @@ define <8 x i16> @test_pshufhw(<8 x i16> %a0, <8 x i16> *%a1) { define <8 x i16> @test_pshuflw(<8 x i16> %a0, <8 x i16> *%a1) { ; GENERIC-LABEL: test_pshuflw: ; GENERIC: # BB#0: -; GENERIC-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,2,4,5,6,7] -; GENERIC-NEXT: pshuflw {{.*#+}} xmm0 = mem[3,2,1,0,4,5,6,7] -; GENERIC-NEXT: paddw %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pshuflw {{.*#+}} xmm1 = xmm0[1,0,3,2,4,5,6,7] sched: [1:0.50] +; GENERIC-NEXT: pshuflw {{.*#+}} xmm0 = mem[3,2,1,0,4,5,6,7] sched: [7:0.50] +; GENERIC-NEXT: paddw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pshuflw: ; ATOM: # BB#0: @@ -4885,10 +4885,10 @@ define <8 x i16> @test_pshuflw(<8 x i16> %a0, <8 x i16> *%a1) { define <4 x i32> @test_pslld(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_pslld: ; GENERIC: # BB#0: -; GENERIC-NEXT: pslld %xmm1, %xmm0 -; GENERIC-NEXT: pslld (%rdi), %xmm0 -; GENERIC-NEXT: pslld $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pslld %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: pslld (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: pslld $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pslld: ; ATOM: # BB#0: @@ -4943,8 +4943,8 @@ declare <4 x i32> @llvm.x86.sse2.pslli.d(<4 x i32>, i32) nounwind readnone define <4 x i32> @test_pslldq(<4 x i32> %a0) { ; GENERIC-LABEL: test_pslldq: ; GENERIC: # BB#0: -; GENERIC-NEXT: pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] -; GENERIC-NEXT: retq +; GENERIC-NEXT: pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11] sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pslldq: ; ATOM: # BB#0: @@ -4988,10 +4988,10 @@ define <4 x i32> @test_pslldq(<4 x i32> %a0) { define <2 x i64> @test_psllq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_psllq: ; GENERIC: # BB#0: -; GENERIC-NEXT: psllq %xmm1, %xmm0 -; GENERIC-NEXT: psllq (%rdi), %xmm0 -; GENERIC-NEXT: psllq $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psllq %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psllq (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psllq $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psllq: ; ATOM: # BB#0: @@ -5046,10 +5046,10 @@ declare <2 x i64> @llvm.x86.sse2.pslli.q(<2 x i64>, i32) nounwind readnone define <8 x i16> @test_psllw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psllw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psllw %xmm1, %xmm0 -; GENERIC-NEXT: psllw (%rdi), %xmm0 -; GENERIC-NEXT: psllw $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psllw %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psllw (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psllw $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psllw: ; ATOM: # BB#0: @@ -5104,10 +5104,10 @@ declare <8 x i16> @llvm.x86.sse2.pslli.w(<8 x i16>, i32) nounwind readnone define <4 x i32> @test_psrad(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_psrad: ; GENERIC: # BB#0: -; GENERIC-NEXT: psrad %xmm1, %xmm0 -; GENERIC-NEXT: psrad (%rdi), %xmm0 -; GENERIC-NEXT: psrad $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psrad %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psrad (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psrad $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psrad: ; ATOM: # BB#0: @@ -5162,10 +5162,10 @@ declare <4 x i32> @llvm.x86.sse2.psrai.d(<4 x i32>, i32) nounwind readnone define <8 x i16> @test_psraw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psraw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psraw %xmm1, %xmm0 -; GENERIC-NEXT: psraw (%rdi), %xmm0 -; GENERIC-NEXT: psraw $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psraw %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psraw (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psraw $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psraw: ; ATOM: # BB#0: @@ -5220,10 +5220,10 @@ declare <8 x i16> @llvm.x86.sse2.psrai.w(<8 x i16>, i32) nounwind readnone define <4 x i32> @test_psrld(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_psrld: ; GENERIC: # BB#0: -; GENERIC-NEXT: psrld %xmm1, %xmm0 -; GENERIC-NEXT: psrld (%rdi), %xmm0 -; GENERIC-NEXT: psrld $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psrld %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psrld (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psrld $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psrld: ; ATOM: # BB#0: @@ -5278,8 +5278,8 @@ declare <4 x i32> @llvm.x86.sse2.psrli.d(<4 x i32>, i32) nounwind readnone define <4 x i32> @test_psrldq(<4 x i32> %a0) { ; GENERIC-LABEL: test_psrldq: ; GENERIC: # BB#0: -; GENERIC-NEXT: psrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero -; GENERIC-NEXT: retq +; GENERIC-NEXT: psrldq {{.*#+}} xmm0 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psrldq: ; ATOM: # BB#0: @@ -5323,10 +5323,10 @@ define <4 x i32> @test_psrldq(<4 x i32> %a0) { define <2 x i64> @test_psrlq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_psrlq: ; GENERIC: # BB#0: -; GENERIC-NEXT: psrlq %xmm1, %xmm0 -; GENERIC-NEXT: psrlq (%rdi), %xmm0 -; GENERIC-NEXT: psrlq $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psrlq %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psrlq (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psrlq $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psrlq: ; ATOM: # BB#0: @@ -5381,10 +5381,10 @@ declare <2 x i64> @llvm.x86.sse2.psrli.q(<2 x i64>, i32) nounwind readnone define <8 x i16> @test_psrlw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psrlw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psrlw %xmm1, %xmm0 -; GENERIC-NEXT: psrlw (%rdi), %xmm0 -; GENERIC-NEXT: psrlw $2, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psrlw %xmm1, %xmm0 # sched: [2:1.00] +; GENERIC-NEXT: psrlw (%rdi), %xmm0 # sched: [8:1.00] +; GENERIC-NEXT: psrlw $2, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psrlw: ; ATOM: # BB#0: @@ -5439,9 +5439,9 @@ declare <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16>, i32) nounwind readnone define <16 x i8> @test_psubb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_psubb: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubb %xmm1, %xmm0 -; GENERIC-NEXT: psubb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubb: ; ATOM: # BB#0: @@ -5491,9 +5491,9 @@ define <16 x i8> @test_psubb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_psubd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_psubd: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubd %xmm1, %xmm0 -; GENERIC-NEXT: psubd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubd (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubd: ; ATOM: # BB#0: @@ -5543,9 +5543,9 @@ define <4 x i32> @test_psubd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <2 x i64> @test_psubq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_psubq: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubq %xmm1, %xmm0 -; GENERIC-NEXT: psubq (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubq (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubq: ; ATOM: # BB#0: @@ -5591,9 +5591,9 @@ define <2 x i64> @test_psubq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <16 x i8> @test_psubsb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_psubsb: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubsb %xmm1, %xmm0 -; GENERIC-NEXT: psubsb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubsb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubsb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubsb: ; ATOM: # BB#0: @@ -5644,9 +5644,9 @@ declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone define <8 x i16> @test_psubsw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psubsw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubsw %xmm1, %xmm0 -; GENERIC-NEXT: psubsw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubsw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubsw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubsw: ; ATOM: # BB#0: @@ -5697,9 +5697,9 @@ declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone define <16 x i8> @test_psubusb(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_psubusb: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubusb %xmm1, %xmm0 -; GENERIC-NEXT: psubusb (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubusb %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubusb (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubusb: ; ATOM: # BB#0: @@ -5750,9 +5750,9 @@ declare <16 x i8> @llvm.x86.sse2.psubus.b(<16 x i8>, <16 x i8>) nounwind readnon define <8 x i16> @test_psubusw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psubusw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubusw %xmm1, %xmm0 -; GENERIC-NEXT: psubusw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubusw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubusw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubusw: ; ATOM: # BB#0: @@ -5803,9 +5803,9 @@ declare <8 x i16> @llvm.x86.sse2.psubus.w(<8 x i16>, <8 x i16>) nounwind readnon define <8 x i16> @test_psubw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_psubw: ; GENERIC: # BB#0: -; GENERIC-NEXT: psubw %xmm1, %xmm0 -; GENERIC-NEXT: psubw (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: psubw %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: psubw (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_psubw: ; ATOM: # BB#0: @@ -5855,9 +5855,9 @@ define <8 x i16> @test_psubw(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <16 x i8> @test_punpckhbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_punpckhbw: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] -; GENERIC-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],mem[8],xmm0[9],mem[9],xmm0[10],mem[10],xmm0[11],mem[11],xmm0[12],mem[12],xmm0[13],mem[13],xmm0[14],mem[14],xmm0[15],mem[15] -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] sched: [1:0.50] +; GENERIC-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8],mem[8],xmm0[9],mem[9],xmm0[10],mem[10],xmm0[11],mem[11],xmm0[12],mem[12],xmm0[13],mem[13],xmm0[14],mem[14],xmm0[15],mem[15] sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpckhbw: ; ATOM: # BB#0: @@ -5907,10 +5907,10 @@ define <16 x i8> @test_punpckhbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_punpckhdq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_punpckhdq: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] -; GENERIC-NEXT: punpckhdq {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50] +; GENERIC-NEXT: punpckhdq {{.*#+}} xmm1 = xmm1[2],mem[2],xmm1[3],mem[3] sched: [7:0.50] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpckhdq: ; ATOM: # BB#0: @@ -5965,10 +5965,10 @@ define <4 x i32> @test_punpckhdq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <2 x i64> @test_punpckhqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_punpckhqdq: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] -; GENERIC-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],mem[1] -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:0.50] +; GENERIC-NEXT: punpckhqdq {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpckhqdq: ; ATOM: # BB#0: @@ -6021,9 +6021,9 @@ define <2 x i64> @test_punpckhqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) define <8 x i16> @test_punpckhwd(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_punpckhwd: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; GENERIC-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50] +; GENERIC-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpckhwd: ; ATOM: # BB#0: @@ -6073,9 +6073,9 @@ define <8 x i16> @test_punpckhwd(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <16 x i8> @test_punpcklbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { ; GENERIC-LABEL: test_punpcklbw: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] -; GENERIC-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] sched: [1:0.50] +; GENERIC-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7] sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpcklbw: ; ATOM: # BB#0: @@ -6125,10 +6125,10 @@ define <16 x i8> @test_punpcklbw(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> *%a2) { define <4 x i32> @test_punpckldq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { ; GENERIC-LABEL: test_punpckldq: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] -; GENERIC-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] -; GENERIC-NEXT: paddd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] sched: [1:0.50] +; GENERIC-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] sched: [7:0.50] +; GENERIC-NEXT: paddd %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpckldq: ; ATOM: # BB#0: @@ -6183,10 +6183,10 @@ define <4 x i32> @test_punpckldq(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> *%a2) { define <2 x i64> @test_punpcklqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_punpcklqdq: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] -; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],mem[0] -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:0.50] +; GENERIC-NEXT: punpcklqdq {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpcklqdq: ; ATOM: # BB#0: @@ -6239,9 +6239,9 @@ define <2 x i64> @test_punpcklqdq(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) define <8 x i16> @test_punpcklwd(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { ; GENERIC-LABEL: test_punpcklwd: ; GENERIC: # BB#0: -; GENERIC-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] -; GENERIC-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] -; GENERIC-NEXT: retq +; GENERIC-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] sched: [1:0.50] +; GENERIC-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3] sched: [7:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_punpcklwd: ; ATOM: # BB#0: @@ -6291,10 +6291,10 @@ define <8 x i16> @test_punpcklwd(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> *%a2) { define <2 x i64> @test_pxor(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { ; GENERIC-LABEL: test_pxor: ; GENERIC: # BB#0: -; GENERIC-NEXT: pxor %xmm1, %xmm0 -; GENERIC-NEXT: pxor (%rdi), %xmm0 -; GENERIC-NEXT: paddq %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: pxor %xmm1, %xmm0 # sched: [1:0.33] +; GENERIC-NEXT: pxor (%rdi), %xmm0 # sched: [7:0.50] +; GENERIC-NEXT: paddq %xmm1, %xmm0 # sched: [1:0.50] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_pxor: ; ATOM: # BB#0: @@ -6347,10 +6347,10 @@ define <2 x i64> @test_pxor(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> *%a2) { define <2 x double> @test_shufpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_shufpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] -; GENERIC-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],mem[0] -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1],xmm1[0] sched: [1:1.00] +; GENERIC-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1],mem[0] sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_shufpd: ; ATOM: # BB#0: @@ -6403,10 +6403,10 @@ define <2 x double> @test_shufpd(<2 x double> %a0, <2 x double> %a1, <2 x double define <2 x double> @test_sqrtpd(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_sqrtpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: sqrtpd %xmm0, %xmm1 -; GENERIC-NEXT: sqrtpd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: sqrtpd %xmm0, %xmm1 # sched: [22:1.00] +; GENERIC-NEXT: sqrtpd (%rdi), %xmm0 # sched: [28:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_sqrtpd: ; ATOM: # BB#0: @@ -6463,11 +6463,11 @@ declare <2 x double> @llvm.x86.sse2.sqrt.pd(<2 x double>) nounwind readnone define <2 x double> @test_sqrtsd(<2 x double> %a0, <2 x double> *%a1) { ; GENERIC-LABEL: test_sqrtsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: sqrtsd %xmm0, %xmm0 -; GENERIC-NEXT: movapd (%rdi), %xmm1 -; GENERIC-NEXT: sqrtsd %xmm1, %xmm1 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: sqrtsd %xmm0, %xmm0 # sched: [22:1.00] +; GENERIC-NEXT: movapd (%rdi), %xmm1 # sched: [6:0.50] +; GENERIC-NEXT: sqrtsd %xmm1, %xmm1 # sched: [22:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_sqrtsd: ; ATOM: # BB#0: @@ -6527,9 +6527,9 @@ declare <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double>) nounwind readnone define <2 x double> @test_subpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_subpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: subpd %xmm1, %xmm0 -; GENERIC-NEXT: subpd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: subpd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_subpd: ; ATOM: # BB#0: @@ -6575,9 +6575,9 @@ define <2 x double> @test_subpd(<2 x double> %a0, <2 x double> %a1, <2 x double> define double @test_subsd(double %a0, double %a1, double *%a2) { ; GENERIC-LABEL: test_subsd: ; GENERIC: # BB#0: -; GENERIC-NEXT: subsd %xmm1, %xmm0 -; GENERIC-NEXT: subsd (%rdi), %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: subsd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: subsd (%rdi), %xmm0 # sched: [9:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_subsd: ; ATOM: # BB#0: @@ -6623,17 +6623,17 @@ define double @test_subsd(double %a0, double %a1, double *%a2) { define i32 @test_ucomisd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_ucomisd: ; GENERIC: # BB#0: -; GENERIC-NEXT: ucomisd %xmm1, %xmm0 -; GENERIC-NEXT: setnp %al -; GENERIC-NEXT: sete %cl -; GENERIC-NEXT: andb %al, %cl -; GENERIC-NEXT: ucomisd (%rdi), %xmm0 -; GENERIC-NEXT: setnp %al -; GENERIC-NEXT: sete %dl -; GENERIC-NEXT: andb %al, %dl -; GENERIC-NEXT: orb %cl, %dl -; GENERIC-NEXT: movzbl %dl, %eax -; GENERIC-NEXT: retq +; GENERIC-NEXT: ucomisd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: setnp %al # sched: [1:1.00] +; GENERIC-NEXT: sete %cl # sched: [1:1.00] +; GENERIC-NEXT: andb %al, %cl # sched: [1:0.33] +; GENERIC-NEXT: ucomisd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: setnp %al # sched: [1:1.00] +; GENERIC-NEXT: sete %dl # sched: [1:1.00] +; GENERIC-NEXT: andb %al, %dl # sched: [1:0.33] +; GENERIC-NEXT: orb %cl, %dl # sched: [1:0.33] +; GENERIC-NEXT: movzbl %dl, %eax # sched: [1:0.33] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_ucomisd: ; ATOM: # BB#0: @@ -6729,10 +6729,10 @@ declare i32 @llvm.x86.sse2.ucomieq.sd(<2 x double>, <2 x double>) nounwind readn define <2 x double> @test_unpckhpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_unpckhpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] -; GENERIC-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] sched: [1:1.00] +; GENERIC-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],mem[1] sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_unpckhpd: ; ATOM: # BB#0: @@ -6785,12 +6785,12 @@ define <2 x double> @test_unpckhpd(<2 x double> %a0, <2 x double> %a1, <2 x doub define <2 x double> @test_unpcklpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_unpcklpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] -; GENERIC-NEXT: movapd %xmm0, %xmm1 -; GENERIC-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],mem[0] -; GENERIC-NEXT: addpd %xmm0, %xmm1 -; GENERIC-NEXT: movapd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0] sched: [1:1.00] +; GENERIC-NEXT: movapd %xmm0, %xmm1 # sched: [1:1.00] +; GENERIC-NEXT: unpcklpd {{.*#+}} xmm1 = xmm1[0],mem[0] sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm0, %xmm1 # sched: [3:1.00] +; GENERIC-NEXT: movapd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_unpcklpd: ; ATOM: # BB#0: @@ -6847,10 +6847,10 @@ define <2 x double> @test_unpcklpd(<2 x double> %a0, <2 x double> %a1, <2 x doub define <2 x double> @test_xorpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) { ; GENERIC-LABEL: test_xorpd: ; GENERIC: # BB#0: -; GENERIC-NEXT: xorpd %xmm1, %xmm0 -; GENERIC-NEXT: xorpd (%rdi), %xmm0 -; GENERIC-NEXT: addpd %xmm1, %xmm0 -; GENERIC-NEXT: retq +; GENERIC-NEXT: xorpd %xmm1, %xmm0 # sched: [1:1.00] +; GENERIC-NEXT: xorpd (%rdi), %xmm0 # sched: [7:1.00] +; GENERIC-NEXT: addpd %xmm1, %xmm0 # sched: [3:1.00] +; GENERIC-NEXT: retq # sched: [1:1.00] ; ; ATOM-LABEL: test_xorpd: ; ATOM: # BB#0: |