[llvm] r329389 - [X86][SandyBridge] Add (V)DPPS memory fold latencies

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Fri Apr  6 04:25:21 2018
New Revision: 329389

URL: http://llvm.org/viewvc/llvm-project?rev=329389&view=rev
Log:
[X86][SandyBridge] Add (V)DPPS memory fold latencies

Noticed this during D44654

Modified:
    llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td
    llvm/trunk/test/CodeGen/X86/avx-schedule.ll
    llvm/trunk/test/CodeGen/X86/sse41-schedule.ll

Modified: llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td?rev=329389&r1=329388&r2=329389&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td (original)
+++ llvm/trunk/lib/Target/X86/X86SchedSandyBridge.td Fri Apr  6 04:25:21 2018
@@ -1954,6 +1954,20 @@ def SBWriteResGroup120 : SchedWriteRes<[
 }
 def: InstRW<[SBWriteResGroup120], (instregex "(V?)DPPDrmi")>;
 
+def SBWriteResGroup121 : SchedWriteRes<[SBPort0,SBPort1,SBPort5,SBPort23]> {
+  let Latency = 18;
+  let NumMicroOps = 5;
+  let ResourceCycles = [1,2,1,1];
+}
+def: InstRW<[SBWriteResGroup121], (instregex "(V?)DPPSrmi")>;
+
+def SBWriteResGroup122 : SchedWriteRes<[SBPort0,SBPort1,SBPort5,SBPort23]> {
+  let Latency = 19;
+  let NumMicroOps = 5;
+  let ResourceCycles = [1,2,1,1];
+}
+def: InstRW<[SBWriteResGroup122], (instregex "VDPPSYrmi")>;
+
 def SBWriteResGroup123 : SchedWriteRes<[SBPort0,SBPort23,SBFPDivider]> {
   let Latency = 20;
   let NumMicroOps = 2;

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=329389&r1=329388&r2=329389&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx-schedule.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx-schedule.ll Fri Apr  6 04:25:21 2018
@@ -1607,13 +1607,13 @@ define <8 x float> @test_dpps(<8 x float
 ; GENERIC-LABEL: test_dpps:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [12:2.00]
-; GENERIC-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
+; GENERIC-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [19:2.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_dpps:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    vdpps $7, %ymm1, %ymm0, %ymm0 # sched: [12:2.00]
-; SANDY-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [8:1.00]
+; SANDY-NEXT:    vdpps $7, (%rdi), %ymm0, %ymm0 # sched: [19:2.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-LABEL: test_dpps:

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=329389&r1=329388&r2=329389&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse41-schedule.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse41-schedule.ll Fri Apr  6 04:25:21 2018
@@ -623,7 +623,7 @@ define <4 x float> @test_dpps(<4 x float
 ; GENERIC-LABEL: test_dpps:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    dpps $7, %xmm1, %xmm0 # sched: [12:2.00]
-; GENERIC-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [8:1.00]
+; GENERIC-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [18:2.00]
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SLM-LABEL: test_dpps:
@@ -635,13 +635,13 @@ define <4 x float> @test_dpps(<4 x float
 ; SANDY-SSE-LABEL: test_dpps:
 ; SANDY-SSE:       # %bb.0:
 ; SANDY-SSE-NEXT:    dpps $7, %xmm1, %xmm0 # sched: [12:2.00]
-; SANDY-SSE-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [8:1.00]
+; SANDY-SSE-NEXT:    dpps $7, (%rdi), %xmm0 # sched: [18:2.00]
 ; SANDY-SSE-NEXT:    retq # sched: [1:1.00]
 ;
 ; SANDY-LABEL: test_dpps:
 ; SANDY:       # %bb.0:
 ; SANDY-NEXT:    vdpps $7, %xmm1, %xmm0, %xmm0 # sched: [12:2.00]
-; SANDY-NEXT:    vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
+; SANDY-NEXT:    vdpps $7, (%rdi), %xmm0, %xmm0 # sched: [18:2.00]
 ; SANDY-NEXT:    retq # sched: [1:1.00]
 ;
 ; HASWELL-SSE-LABEL: test_dpps: