[llvm] r320308 - [X86] Flag BroadWell scheduler model as complete

Sanjoy Das via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 12 10:51:18 PST 2017


On Tue, Dec 12, 2017 at 10:47 AM, Craig Topper <craig.topper at gmail.com> wrote:
> Simon already reverted the important part that makes this an error in
> r320492.

Apologies; I assumed that because git-svnrevert succeeded nothing must
have changed in the meantime.  I'll revert the revert.

-- Sanjoy

>
> On Tue, Dec 12, 2017 at 10:43 AM Sanjoy Das via llvm-commits
> <llvm-commits at lists.llvm.org> wrote:
>>
>> Hi Simon,
>>
>> I've reverted this commit in r320508
>>
>> -- Sanjoy
>>
>> On Mon, Dec 11, 2017 at 5:56 PM, Sanjoy Das
>> <sanjoy at playingwithpointers.com> wrote:
>> > Hi Simon,
>> >
>> > This commit breaks XLA:CPU -- the attached IR crashes LLVM on trunk
>> > when compiled as `llc -O3 -mcpu=broadwell bug.ll` after this change.
>> > Can you please fix or revert?
>> >
>> > Here is the backtrace:
>> >
>> > DefIdx 1 exceeds machine model writes for %53:gr64, %5:gr64 = MULX64rr
>> > %114, implicit killed %rdx; GR64:%53,%5,%114
>> >  (Try with MCSchedModel.CompleteModel set to false)incomplete machine
>> > model
>> > UNREACHABLE executed at ../../lib/CodeGen/TargetSchedule.cpp:249!
>> >
>> > Program received signal SIGABRT, Aborted.
>> > 0x00007ffff67bbc37 in __GI_raise (sig=sig at entry=6) at
>> > ../nptl/sysdeps/unix/sysv/linux/raise.c:56
>> > 56      ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or
>> > directory.
>> > (gdb) bt
>> > #0  0x00007ffff67bbc37 in __GI_raise (sig=sig at entry=6) at
>> > ../nptl/sysdeps/unix/sysv/linux/raise.c:56
>> > #1  0x00007ffff67bf028 in __GI_abort () at abort.c:89
>> > #2  0x0000000002f3a370 in llvm::llvm_unreachable_internal
>> > (msg=0x7dccdb "incomplete machine model", file=0x7dcbed
>> > "../../lib/CodeGen/TargetSchedule.cpp", line=249)
>> >     at ../../lib/Support/ErrorHandling.cpp:189
>> > #3  0x00000000024a7654 in
>> > llvm::TargetSchedModel::computeOperandLatency (this=0x39764d0,
>> > DefMI=0x393cdb0, DefOperIdx=1, UseMI=0x393d0b0, UseOperIdx=1)
>> >     at ../../lib/CodeGen/TargetSchedule.cpp:249
>> > #4  0x00000000023e8da7 in llvm::ScheduleDAGInstrs::addVRegDefDeps
>> > (this=0x3976230, SU=0x3978060, OperIdx=1) at
>> > ../../lib/CodeGen/ScheduleDAGInstrs.cpp:409
>> > #5  0x00000000023ea4ea in llvm::ScheduleDAGInstrs::buildSchedGraph
>> > (this=0x3976230, AA=0x3930880, RPTracker=0x3976d70, PDiffs=0x3976c80,
>> > LIS=0x38ab400, TrackLaneMasks=false)
>> >     at ../../lib/CodeGen/ScheduleDAGInstrs.cpp:803
>> > #6  0x0000000002270331 in
>> > llvm::ScheduleDAGMILive::buildDAGWithRegPressure (this=0x3976230) at
>> > ../../lib/CodeGen/MachineScheduler.cpp:1288
>> > #7  0x000000000226fa1e in llvm::ScheduleDAGMILive::schedule
>> > (this=0x3976230) at ../../lib/CodeGen/MachineScheduler.cpp:1194
>> > #8  0x0000000002279feb in (anonymous
>> > namespace)::MachineSchedulerBase::scheduleRegions (this=0x38abdf0,
>> > Scheduler=..., FixKillFlags=false)
>> >     at ../../lib/CodeGen/MachineScheduler.cpp:564
>> > #9  0x0000000002279495 in (anonymous
>> > namespace)::MachineScheduler::runOnMachineFunction (this=0x38abdf0,
>> > mf=...) at ../../lib/CodeGen/MachineScheduler.cpp:383
>> > #10 0x000000000216b401 in llvm::MachineFunctionPass::runOnFunction
>> > (this=0x38abe30, F=...) at
>> > ../../lib/CodeGen/MachineFunctionPass.cpp:62
>> > #11 0x000000000267b46f in llvm::FPPassManager::runOnFunction
>> > (this=0x3896a80, F=...) at ../../lib/IR/LegacyPassManager.cpp:1520
>> > #12 0x000000000267b785 in llvm::FPPassManager::runOnModule
>> > (this=0x3896a80, M=...) at ../../lib/IR/LegacyPassManager.cpp:1541
>> > #13 0x000000000267bf1a in (anonymous
>> > namespace)::MPPassManager::runOnModule (this=0x387bf60, M=...) at
>> > ../../lib/IR/LegacyPassManager.cpp:1597
>> > #14 0x000000000267ba46 in llvm::legacy::PassManagerImpl::run
>> > (this=0x387d780, M=...) at ../../lib/IR/LegacyPassManager.cpp:1700
>> > #15 0x000000000267c431 in llvm::legacy::PassManager::run
>> > (this=0x7fffffffd7d0, M=...) at
>> > ../../lib/IR/LegacyPassManager.cpp:1731
>> > #16 0x00000000017e721d in compileModule (argv=0x7fffffffdec8,
>> > Context=...) at ../../tools/llc/llc.cpp:569
>> > #17 0x00000000017e587a in main (argc=4, argv=0x7fffffffdec8) at
>> > ../../tools/llc/llc.cpp:346
>> >
>> >
>> > -- Sanjoy
>> >
>> > On Sun, Dec 10, 2017 at 5:49 AM, Simon Pilgrim via llvm-commits
>> > <llvm-commits at lists.llvm.org> wrote:
>> >> Author: rksimon
>> >> Date: Sun Dec 10 05:49:51 2017
>> >> New Revision: 320308
>> >>
>> >> URL: http://llvm.org/viewvc/llvm-project?rev=320308&view=rev
>> >> Log:
>> >> [X86] Flag BroadWell scheduler model as complete
>> >>
>> >> Locally tag COPY as WriteMove, which has caused some reg-reg + reg-mem
>> >> instruction tests to reorder.
>> >>
>> >> Modified:
>> >>     llvm/trunk/lib/Target/X86/X86SchedBroadwell.td
>> >>     llvm/trunk/test/CodeGen/X86/aes-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/avx-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/avx2-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/mmx-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/sse-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/sse2-schedule.ll
>> >>     llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
>> >>
>> >> Modified: llvm/trunk/lib/Target/X86/X86SchedBroadwell.td
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86SchedBroadwell.td?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/lib/Target/X86/X86SchedBroadwell.td (original)
>> >> +++ llvm/trunk/lib/Target/X86/X86SchedBroadwell.td Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -21,10 +21,6 @@ def BroadwellModel : SchedMachineModel {
>> >>
>> >>    // Based on the LSD (loop-stream detector) queue size and
>> >> benchmarking data.
>> >>    let LoopMicroOpBufferSize = 50;
>> >> -
>> >> -  // This flag is set to allow the scheduler to assign a default model
>> >> to
>> >> -  // unrecognized opcodes.
>> >> -  let CompleteModel = 0;
>> >>  }
>> >>
>> >>  let SchedModel = BroadwellModel in {
>> >> @@ -120,6 +116,9 @@ def : WriteRes<WriteMove,  [BWPort0156]>
>> >>  // These can often bypass execution ports completely.
>> >>  def : WriteRes<WriteZero,  []>;
>> >>
>> >> +// Treat misc copies as a move.
>> >> +def : InstRW<[WriteMove], (instrs COPY)>;
>> >> +
>> >>  // Branches don't produce values, so they have no latency, but they
>> >> still
>> >>  // consume resources. Indirect branches can fold loads.
>> >>  defm : BWWriteResPair<WriteJump,  BWPort06,   1>;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/aes-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/aes-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/aes-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/aes-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -261,8 +261,8 @@ define <2 x i64> @test_aesimc(<2 x i64>
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_aesimc:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vaesimc %xmm0, %xmm0 # sched: [14:2.00]
>> >>  ; BROADWELL-NEXT:    vaesimc (%rdi), %xmm1 # sched: [19:2.00]
>> >> +; BROADWELL-NEXT:    vaesimc %xmm0, %xmm0 # sched: [14:2.00]
>> >>  ; BROADWELL-NEXT:    vpor %xmm1, %xmm0, %xmm0 # sched: [1:0.33]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/avx-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/avx-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/avx-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -1069,8 +1069,8 @@ define <4 x double> @test_cvtdq2pd(<4 x
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtdq2pd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtdq2pd %xmm0, %ymm0 # sched: [6:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtdq2pd (%rdi), %ymm1 # sched: [11:1.00]
>> >> +; BROADWELL-NEXT:    vcvtdq2pd %xmm0, %ymm0 # sched: [6:1.00]
>> >>  ; BROADWELL-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4035,8 +4035,8 @@ define <8 x float> @test_rcpps(<8 x floa
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_rcpps:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vrcpps (%rdi), %ymm1 # sched: [17:2.00]
>> >>  ; BROADWELL-NEXT:    vrcpps %ymm0, %ymm0 # sched: [11:2.00]
>> >> +; BROADWELL-NEXT:    vrcpps (%rdi), %ymm1 # sched: [17:2.00]
>> >>  ; BROADWELL-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4099,8 +4099,8 @@ define <4 x double> @test_roundpd(<4 x d
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundpd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundpd $7, %ymm0, %ymm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vroundpd $7, (%rdi), %ymm1 # sched: [12:2.00]
>> >> +; BROADWELL-NEXT:    vroundpd $7, %ymm0, %ymm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4163,8 +4163,8 @@ define <8 x float> @test_roundps(<8 x fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundps:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundps $7, %ymm0, %ymm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vroundps $7, (%rdi), %ymm1 # sched: [12:2.00]
>> >> +; BROADWELL-NEXT:    vroundps $7, %ymm0, %ymm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4227,8 +4227,8 @@ define <8 x float> @test_rsqrtps(<8 x fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_rsqrtps:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vrsqrtps (%rdi), %ymm1 # sched: [17:2.00]
>> >>  ; BROADWELL-NEXT:    vrsqrtps %ymm0, %ymm0 # sched: [11:2.00]
>> >> +; BROADWELL-NEXT:    vrsqrtps (%rdi), %ymm1 # sched: [17:2.00]
>> >>  ; BROADWELL-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4408,8 +4408,8 @@ define <4 x double> @test_sqrtpd(<4 x do
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_sqrtpd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vsqrtpd (%rdi), %ymm1 # sched: [40:2.00]
>> >>  ; BROADWELL-NEXT:    vsqrtpd %ymm0, %ymm0 # sched: [34:2.00]
>> >> +; BROADWELL-NEXT:    vsqrtpd (%rdi), %ymm1 # sched: [40:2.00]
>> >>  ; BROADWELL-NEXT:    vaddpd %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -4472,8 +4472,8 @@ define <8 x float> @test_sqrtps(<8 x flo
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_sqrtps:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vsqrtps (%rdi), %ymm1 # sched: [27:2.00]
>> >>  ; BROADWELL-NEXT:    vsqrtps %ymm0, %ymm0 # sched: [21:2.00]
>> >> +; BROADWELL-NEXT:    vsqrtps (%rdi), %ymm1 # sched: [27:2.00]
>> >>  ; BROADWELL-NEXT:    vaddps %ymm1, %ymm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/avx2-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx2-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/avx2-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/avx2-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -1761,8 +1761,8 @@ define <16 x i8> @test_pbroadcastb(<16 x
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_pbroadcastb:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vpbroadcastb %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpbroadcastb (%rdi), %xmm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vpbroadcastb %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpaddb %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1810,8 +1810,8 @@ define <32 x i8> @test_pbroadcastb_ymm(<
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_pbroadcastb_ymm:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vpbroadcastb %xmm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpbroadcastb (%rdi), %ymm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vpbroadcastb %xmm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpaddb %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -2051,8 +2051,8 @@ define <8 x i16> @test_pbroadcastw(<8 x
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_pbroadcastw:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vpbroadcastw %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpbroadcastw (%rdi), %xmm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vpbroadcastw %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpaddw %xmm1, %xmm0, %xmm0 # sched: [1:0.50]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -2100,8 +2100,8 @@ define <16 x i16> @test_pbroadcastw_ymm(
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_pbroadcastw_ymm:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vpbroadcastw %xmm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpbroadcastw (%rdi), %ymm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vpbroadcastw %xmm0, %ymm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    vpaddw %ymm1, %ymm0, %ymm0 # sched: [1:0.50]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/mmx-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/mmx-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/mmx-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/mmx-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -54,10 +54,10 @@ define i64 @test_cvtpd2pi(<2 x double> %
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtpd2pi:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    cvtpd2pi (%rdi), %mm0 # sched: [9:1.00]
>> >> -; BROADWELL-NEXT:    cvtpd2pi %xmm0, %mm1 # sched: [4:1.00]
>> >> -; BROADWELL-NEXT:    por %mm1, %mm0 # sched: [1:0.33]
>> >> -; BROADWELL-NEXT:    movd %mm0, %rax # sched: [1:1.00]
>> >> +; BROADWELL-NEXT:    cvtpd2pi %xmm0, %mm0 # sched: [4:1.00]
>> >> +; BROADWELL-NEXT:    cvtpd2pi (%rdi), %mm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    por %mm0, %mm1 # sched: [1:0.33]
>> >> +; BROADWELL-NEXT:    movd %mm1, %rax # sched: [1:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>  ; SKYLAKE-LABEL: test_cvtpd2pi:
>> >> @@ -138,9 +138,9 @@ define <2 x double> @test_cvtpi2pd(x86_m
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtpi2pd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    cvtpi2pd %mm0, %xmm0 # sched: [4:1.00]
>> >> -; BROADWELL-NEXT:    cvtpi2pd (%rdi), %xmm1 # sched: [9:1.00]
>> >> -; BROADWELL-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
>> >> +; BROADWELL-NEXT:    cvtpi2pd (%rdi), %xmm0 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    cvtpi2pd %mm0, %xmm1 # sched: [4:1.00]
>> >> +; BROADWELL-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>  ; SKYLAKE-LABEL: test_cvtpi2pd:
>> >> @@ -388,10 +388,10 @@ define i64 @test_cvttpd2pi(<2 x double>
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvttpd2pi:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    cvttpd2pi (%rdi), %mm0 # sched: [9:1.00]
>> >> -; BROADWELL-NEXT:    cvttpd2pi %xmm0, %mm1 # sched: [4:1.00]
>> >> -; BROADWELL-NEXT:    por %mm1, %mm0 # sched: [1:0.33]
>> >> -; BROADWELL-NEXT:    movd %mm0, %rax # sched: [1:1.00]
>> >> +; BROADWELL-NEXT:    cvttpd2pi %xmm0, %mm0 # sched: [4:1.00]
>> >> +; BROADWELL-NEXT:    cvttpd2pi (%rdi), %mm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    por %mm0, %mm1 # sched: [1:0.33]
>> >> +; BROADWELL-NEXT:    movd %mm1, %rax # sched: [1:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>  ; SKYLAKE-LABEL: test_cvttpd2pi:
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/sse-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/sse-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/sse-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -784,8 +784,8 @@ define i32 @test_cvtss2si(float %a0, flo
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtss2si:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtss2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtss2si (%rdi), %eax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvtss2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addl %ecx, %eax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -864,8 +864,8 @@ define i64 @test_cvtss2siq(float %a0, fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtss2siq:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtss2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtss2si (%rdi), %rax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvtss2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addq %rcx, %rax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -944,8 +944,8 @@ define i32 @test_cvttss2si(float %a0, fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvttss2si:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvttss2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvttss2si (%rdi), %eax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvttss2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addl %ecx, %eax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1021,8 +1021,8 @@ define i64 @test_cvttss2siq(float %a0, f
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvttss2siq:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvttss2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvttss2si (%rdi), %rax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvttss2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addq %rcx, %rax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/sse2-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/sse2-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/sse2-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -705,8 +705,8 @@ define <2 x double> @test_cvtdq2pd(<4 x
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtdq2pd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtdq2pd %xmm0, %xmm0 # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtdq2pd (%rdi), %xmm1 # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvtdq2pd %xmm0, %xmm0 # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1179,8 +1179,8 @@ define i32 @test_cvtsd2si(double %a0, do
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtsd2si:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtsd2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtsd2si (%rdi), %eax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvtsd2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addl %ecx, %eax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1259,8 +1259,8 @@ define i64 @test_cvtsd2siq(double %a0, d
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvtsd2siq:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvtsd2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvtsd2si (%rdi), %rax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvtsd2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addq %rcx, %rax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1829,8 +1829,8 @@ define i32 @test_cvttsd2si(double %a0, d
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvttsd2si:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvttsd2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvttsd2si (%rdi), %eax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvttsd2si %xmm0, %ecx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addl %ecx, %eax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -1906,8 +1906,8 @@ define i64 @test_cvttsd2siq(double %a0,
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_cvttsd2siq:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vcvttsd2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    vcvttsd2si (%rdi), %rax # sched: [9:1.00]
>> >> +; BROADWELL-NEXT:    vcvttsd2si %xmm0, %rcx # sched: [4:1.00]
>> >>  ; BROADWELL-NEXT:    addq %rcx, %rax # sched: [1:0.25]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>
>> >> Modified: llvm/trunk/test/CodeGen/X86/sse41-schedule.ll
>> >> URL:
>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse41-schedule.ll?rev=320308&r1=320307&r2=320308&view=diff
>> >>
>> >> ==============================================================================
>> >> --- llvm/trunk/test/CodeGen/X86/sse41-schedule.ll (original)
>> >> +++ llvm/trunk/test/CodeGen/X86/sse41-schedule.ll Sun Dec 10 05:49:51
>> >> 2017
>> >> @@ -2992,8 +2992,8 @@ define <2 x double> @test_roundpd(<2 x d
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundpd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundpd $7, %xmm0, %xmm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vroundpd $7, (%rdi), %xmm1 # sched: [11:2.00]
>> >> +; BROADWELL-NEXT:    vroundpd $7, %xmm0, %xmm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -3064,8 +3064,8 @@ define <4 x float> @test_roundps(<4 x fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundps:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundps $7, %xmm0, %xmm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vroundps $7, (%rdi), %xmm1 # sched: [11:2.00]
>> >> +; BROADWELL-NEXT:    vroundps $7, %xmm0, %xmm0 # sched: [6:0.50]
>> >>  ; BROADWELL-NEXT:    vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >> @@ -3137,9 +3137,9 @@ define <2 x double> @test_roundsd(<2 x d
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundsd:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundsd $7, %xmm1, %xmm0, %xmm1 # sched:
>> >> [6:0.50]
>> >> -; BROADWELL-NEXT:    vroundsd $7, (%rdi), %xmm0, %xmm0 # sched:
>> >> [11:2.00]
>> >> -; BROADWELL-NEXT:    vaddpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
>> >> +; BROADWELL-NEXT:    vroundsd $7, (%rdi), %xmm0, %xmm2 # sched:
>> >> [11:2.00]
>> >> +; BROADWELL-NEXT:    vroundsd $7, %xmm1, %xmm0, %xmm0 # sched:
>> >> [6:0.50]
>> >> +; BROADWELL-NEXT:    vaddpd %xmm2, %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>  ; SKYLAKE-LABEL: test_roundsd:
>> >> @@ -3210,9 +3210,9 @@ define <4 x float> @test_roundss(<4 x fl
>> >>  ;
>> >>  ; BROADWELL-LABEL: test_roundss:
>> >>  ; BROADWELL:       # %bb.0:
>> >> -; BROADWELL-NEXT:    vroundss $7, %xmm1, %xmm0, %xmm1 # sched:
>> >> [6:0.50]
>> >> -; BROADWELL-NEXT:    vroundss $7, (%rdi), %xmm0, %xmm0 # sched:
>> >> [11:2.00]
>> >> -; BROADWELL-NEXT:    vaddps %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
>> >> +; BROADWELL-NEXT:    vroundss $7, (%rdi), %xmm0, %xmm2 # sched:
>> >> [11:2.00]
>> >> +; BROADWELL-NEXT:    vroundss $7, %xmm1, %xmm0, %xmm0 # sched:
>> >> [6:0.50]
>> >> +; BROADWELL-NEXT:    vaddps %xmm2, %xmm0, %xmm0 # sched: [3:1.00]
>> >>  ; BROADWELL-NEXT:    retq # sched: [7:1.00]
>> >>  ;
>> >>  ; SKYLAKE-LABEL: test_roundss:
>> >>
>> >>
>> >> _______________________________________________
>> >> llvm-commits mailing list
>> >> llvm-commits at lists.llvm.org
>> >> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
> --
> ~Craig


More information about the llvm-commits mailing list