[X86] Fix scheduler class for EVEX VMOVNTDQA variants and cleanup Skylake/Icelake resource usage #116946
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…lake/Icelake resource usage Ensure we use SchedWriteVecMoveLSNT for (V)MOVNTDQA instructions, remove unnecessary scheduler overrides and adjust resource pipe usage to match uops.info/Agner numbers
|
@llvm/pr-subscribers-backend-x86 Author: Simon Pilgrim (RKSimon) ChangesEnsure we use the SchedWriteVecMoveLSNT class for all (V)MOVNTDQA instructions, remove unnecessary scheduler overrides and adjust resource pipe usage to match uops.info/Agner numbers Patch is 35.10 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/116946.diff 15 Files Affected:
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 7f854309eeed04..a05a3063cac559 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -4549,20 +4549,20 @@ let Predicates = [HasAVX512] in {
def VMOVNTDQAZrm : AVX512PI<0x2A, MRMSrcMem, (outs VR512:$dst),
(ins i512mem:$src), "vmovntdqa\t{$src, $dst|$dst, $src}",
- [], SSEPackedInt>, Sched<[SchedWriteVecMoveLS.ZMM.RM]>,
+ [], SSEPackedInt>, Sched<[SchedWriteVecMoveLSNT.ZMM.RM]>,
EVEX, T8, PD, EVEX_V512, EVEX_CD8<64, CD8VF>;
let Predicates = [HasVLX] in {
def VMOVNTDQAZ256rm : AVX512PI<0x2A, MRMSrcMem, (outs VR256X:$dst),
(ins i256mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
- [], SSEPackedInt>, Sched<[SchedWriteVecMoveLS.YMM.RM]>,
+ [], SSEPackedInt>, Sched<[SchedWriteVecMoveLSNT.YMM.RM]>,
EVEX, T8, PD, EVEX_V256, EVEX_CD8<64, CD8VF>;
def VMOVNTDQAZ128rm : AVX512PI<0x2A, MRMSrcMem, (outs VR128X:$dst),
(ins i128mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
- [], SSEPackedInt>, Sched<[SchedWriteVecMoveLS.XMM.RM]>,
+ [], SSEPackedInt>, Sched<[SchedWriteVecMoveLSNT.XMM.RM]>,
EVEX, T8, PD, EVEX_V128, EVEX_CD8<64, CD8VF>;
}
diff --git a/llvm/lib/Target/X86/X86SchedIceLake.td b/llvm/lib/Target/X86/X86SchedIceLake.td
index 48be485176c3fa..4bc9f76e6f2c4c 100644
--- a/llvm/lib/Target/X86/X86SchedIceLake.td
+++ b/llvm/lib/Target/X86/X86SchedIceLake.td
@@ -350,8 +350,8 @@ defm : ICXWriteResPair<WriteFVarBlendZ,[ICXPort015], 2, [2], 2, 7>;
defm : X86WriteRes<WriteVecLoad, [ICXPort23], 5, [1], 1>;
defm : X86WriteRes<WriteVecLoadX, [ICXPort23], 6, [1], 1>;
defm : X86WriteRes<WriteVecLoadY, [ICXPort23], 7, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNT, [ICXPort23], 6, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNTY, [ICXPort23], 7, [1], 1>;
+defm : X86WriteRes<WriteVecLoadNT, [ICXPort23,ICXPort015], 7, [1,1], 2>;
+defm : X86WriteRes<WriteVecLoadNTY, [ICXPort23,ICXPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoad, [ICXPort23,ICXPort015], 7, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoadY, [ICXPort23,ICXPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecStore, [ICXPort78,ICXPort49], 1, [1,1], 2>;
@@ -1361,8 +1361,7 @@ def ICXWriteResGroup95 : SchedWriteRes<[ICXPort23,ICXPort015]> {
let NumMicroOps = 2;
let ReleaseAtCycles = [1,1];
}
-def: InstRW<[ICXWriteResGroup95], (instrs VMOVNTDQAZ128rm,
- VPBLENDDrmi)>;
+def: InstRW<[ICXWriteResGroup95], (instrs VPBLENDDrmi)>;
def: InstRW<[ICXWriteResGroup95, ReadAfterVecXLd],
(instregex "VBLENDMPDZ128rm(b?)",
"VBLENDMPSZ128rm(b?)",
@@ -1568,8 +1567,7 @@ def ICXWriteResGroup121 : SchedWriteRes<[ICXPort23,ICXPort015]> {
let NumMicroOps = 2;
let ReleaseAtCycles = [1,1];
}
-def: InstRW<[ICXWriteResGroup121], (instrs VMOVNTDQAZ256rm,
- VPBLENDDYrmi)>;
+def: InstRW<[ICXWriteResGroup121], (instrs VPBLENDDYrmi)>;
def: InstRW<[ICXWriteResGroup121, ReadAfterVecYLd],
(instregex "VBLENDMPD(Z|Z256)rm(b?)",
"VBLENDMPS(Z|Z256)rm(b?)",
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeClient.td b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
index 116aa3555a065c..7652eb7325ac1c 100644
--- a/llvm/lib/Target/X86/X86SchedSkylakeClient.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeClient.td
@@ -345,8 +345,8 @@ defm : X86WriteResPairUnsupported<WriteFVarBlendZ>;
defm : X86WriteRes<WriteVecLoad, [SKLPort23], 5, [1], 1>;
defm : X86WriteRes<WriteVecLoadX, [SKLPort23], 6, [1], 1>;
defm : X86WriteRes<WriteVecLoadY, [SKLPort23], 7, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNT, [SKLPort23], 6, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNTY, [SKLPort23], 7, [1], 1>;
+defm : X86WriteRes<WriteVecLoadNT, [SKLPort23,SKLPort015], 7, [1,1], 2>;
+defm : X86WriteRes<WriteVecLoadNTY, [SKLPort23,SKLPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoad, [SKLPort23,SKLPort015], 7, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoadY, [SKLPort23,SKLPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecStore, [SKLPort237,SKLPort4], 1, [1,1], 2>;
diff --git a/llvm/lib/Target/X86/X86SchedSkylakeServer.td b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
index 550edaa05f7baa..038650c6fd74ac 100644
--- a/llvm/lib/Target/X86/X86SchedSkylakeServer.td
+++ b/llvm/lib/Target/X86/X86SchedSkylakeServer.td
@@ -345,8 +345,8 @@ defm : SKXWriteResPair<WriteFVarBlendZ,[SKXPort015], 2, [2], 2, 7>;
defm : X86WriteRes<WriteVecLoad, [SKXPort23], 5, [1], 1>;
defm : X86WriteRes<WriteVecLoadX, [SKXPort23], 6, [1], 1>;
defm : X86WriteRes<WriteVecLoadY, [SKXPort23], 7, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNT, [SKXPort23], 6, [1], 1>;
-defm : X86WriteRes<WriteVecLoadNTY, [SKXPort23], 7, [1], 1>;
+defm : X86WriteRes<WriteVecLoadNT, [SKXPort23,SKXPort015], 7, [1,1], 2>;
+defm : X86WriteRes<WriteVecLoadNTY, [SKXPort23,SKXPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoad, [SKXPort23,SKXPort015], 7, [1,1], 2>;
defm : X86WriteRes<WriteVecMaskedLoadY, [SKXPort23,SKXPort015], 8, [1,1], 2>;
defm : X86WriteRes<WriteVecStore, [SKXPort237,SKXPort4], 1, [1,1], 2>;
@@ -1336,8 +1336,7 @@ def SKXWriteResGroup95 : SchedWriteRes<[SKXPort23,SKXPort015]> {
let NumMicroOps = 2;
let ReleaseAtCycles = [1,1];
}
-def: InstRW<[SKXWriteResGroup95], (instrs VMOVNTDQAZ128rm,
- VPBLENDDrmi)>;
+def: InstRW<[SKXWriteResGroup95], (instrs VPBLENDDrmi)>;
def: InstRW<[SKXWriteResGroup95, ReadAfterVecXLd],
(instregex "VBLENDMPDZ128rm(b?)",
"VBLENDMPSZ128rm(b?)",
@@ -1539,8 +1538,7 @@ def SKXWriteResGroup121 : SchedWriteRes<[SKXPort23,SKXPort015]> {
let NumMicroOps = 2;
let ReleaseAtCycles = [1,1];
}
-def: InstRW<[SKXWriteResGroup121], (instrs VMOVNTDQAZ256rm,
- VPBLENDDYrmi)>;
+def: InstRW<[SKXWriteResGroup121], (instrs VPBLENDDYrmi)>;
def: InstRW<[SKXWriteResGroup121, ReadAfterVecYLd],
(instregex "VBLENDMPD(Z|Z256)rm(b?)",
"VBLENDMPS(Z|Z256)rm(b?)",
diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx1.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx1.s
index edd3e7cd8c0e79..0e7c5751e5c084 100644
--- a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx1.s
+++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx1.s
@@ -1300,8 +1300,8 @@ vzeroupper
# CHECK-NEXT: 1 2 1.00 vmovmskps %ymm0, %ecx
# CHECK-NEXT: 2 1 0.50 * vmovntdq %xmm0, (%rax)
# CHECK-NEXT: 2 1 0.50 * vmovntdq %ymm0, (%rax)
-# CHECK-NEXT: 1 6 0.50 * vmovntdqa (%rax), %xmm2
-# CHECK-NEXT: 1 7 0.50 * vmovntdqa (%rax), %ymm2
+# CHECK-NEXT: 2 7 0.50 * vmovntdqa (%rax), %xmm2
+# CHECK-NEXT: 2 8 0.50 * vmovntdqa (%rax), %ymm2
# CHECK-NEXT: 2 1 0.50 * vmovntpd %xmm0, (%rax)
# CHECK-NEXT: 2 1 0.50 * vmovntpd %ymm0, (%rax)
# CHECK-NEXT: 2 1 0.50 * vmovntps %xmm0, (%rax)
@@ -1738,7 +1738,7 @@ vzeroupper
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
-# CHECK-NEXT: - 126.00 325.58 256.58 160.50 160.50 19.00 270.58 6.25 19.00 19.00 19.00
+# CHECK-NEXT: - 126.00 326.25 257.25 160.50 160.50 19.00 271.25 6.25 19.00 19.00 19.00
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
@@ -2012,8 +2012,8 @@ vzeroupper
# CHECK-NEXT: - - 1.00 - - - - - - - - - vmovmskps %ymm0, %ecx
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovntdq %xmm0, (%rax)
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovntdq %ymm0, (%rax)
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovntdqa (%rax), %xmm2
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovntdqa (%rax), %ymm2
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovntdqa (%rax), %xmm2
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovntdqa (%rax), %ymm2
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovntpd %xmm0, (%rax)
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovntpd %ymm0, (%rax)
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovntps %xmm0, (%rax)
diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx2.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx2.s
index 81b178d12d2d33..c7a0be0cf9cde6 100644
--- a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx2.s
+++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx2.s
@@ -475,7 +475,7 @@ vpxor (%rax), %ymm1, %ymm2
# CHECK-NEXT: 5 20 2.00 * vgatherqps %xmm0, (%rax,%ymm1,2), %xmm2
# CHECK-NEXT: 1 3 1.00 vinserti128 $1, %xmm0, %ymm1, %ymm2
# CHECK-NEXT: 2 7 0.50 * vinserti128 $1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT: 1 7 0.50 * vmovntdqa (%rax), %ymm0
+# CHECK-NEXT: 2 8 0.50 * vmovntdqa (%rax), %ymm0
# CHECK-NEXT: 2 4 1.00 vmpsadbw $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: 3 11 1.00 * vmpsadbw $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: 1 1 0.50 vpabsb %ymm0, %ymm2
@@ -778,7 +778,7 @@ vpxor (%rax), %ymm1, %ymm2
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
-# CHECK-NEXT: - - 110.33 116.33 98.00 98.00 2.50 137.33 - 2.50 2.50 2.50
+# CHECK-NEXT: - - 110.67 116.67 98.00 98.00 2.50 137.67 - 2.50 2.50 2.50
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
@@ -797,7 +797,7 @@ vpxor (%rax), %ymm1, %ymm2
# CHECK-NEXT: - - 1.33 0.33 2.00 2.00 - 1.33 - - - - vgatherqps %xmm0, (%rax,%ymm1,2), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - vinserti128 $1, %xmm0, %ymm1, %ymm2
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vinserti128 $1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovntdqa (%rax), %ymm0
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovntdqa (%rax), %ymm0
# CHECK-NEXT: - - - 0.50 - - - 1.50 - - - - vmpsadbw $1, %ymm0, %ymm1, %ymm2
# CHECK-NEXT: - - - 0.50 0.50 0.50 - 1.50 - - - - vmpsadbw $1, (%rax), %ymm1, %ymm2
# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpabsb %ymm0, %ymm2
diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512.s
index 28abdd5de1ce97..8b495d6ee268e5 100644
--- a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512.s
+++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512.s
@@ -1499,7 +1499,7 @@ vunpcklps (%rax){1to16}, %zmm17, %zmm19 {z}{k1}
# CHECK-NEXT: 2 1 0.50 * vmovdqu64 %zmm16, (%rax) {%k1}
# CHECK-NEXT: 1 1 0.50 vmovdqu64 %zmm16, %zmm19 {%k1} {z}
# CHECK-NEXT: 2 8 0.50 * vmovdqu64 (%rax), %zmm19 {%k1} {z}
-# CHECK-NEXT: 1 7 0.50 * vmovntdqa (%rax), %zmm0
+# CHECK-NEXT: 2 8 0.50 * vmovntdqa (%rax), %zmm0
# CHECK-NEXT: 1 1 1.00 vmovshdup %zmm16, %zmm19
# CHECK-NEXT: 2 8 0.50 * vmovshdup (%rax), %zmm19
# CHECK-NEXT: 1 1 1.00 vmovshdup %zmm16, %zmm19 {%k1}
@@ -2057,7 +2057,7 @@ vunpcklps (%rax){1to16}, %zmm17, %zmm19 {z}{k1}
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
-# CHECK-NEXT: - 612.00 411.17 103.67 328.00 328.00 48.50 593.17 6.00 48.50 48.50 48.50
+# CHECK-NEXT: - 612.00 411.50 104.00 328.00 328.00 48.50 593.50 6.00 48.50 48.50 48.50
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
@@ -2470,7 +2470,7 @@ vunpcklps (%rax){1to16}, %zmm17, %zmm19 {z}{k1}
# CHECK-NEXT: - - - - - - 0.50 - - 0.50 0.50 0.50 vmovdqu64 %zmm16, (%rax) {%k1}
# CHECK-NEXT: - - 0.50 - - - - 0.50 - - - - vmovdqu64 %zmm16, %zmm19 {%k1} {z}
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovdqu64 (%rax), %zmm19 {%k1} {z}
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - vmovntdqa (%rax), %zmm0
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovntdqa (%rax), %zmm0
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %zmm16, %zmm19
# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - vmovshdup (%rax), %zmm19
# CHECK-NEXT: - - - - - - - 1.00 - - - - vmovshdup %zmm16, %zmm19 {%k1}
diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse41.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse41.s
index 6ff620b0779f3c..9748a278487704 100644
--- a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse41.s
+++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-sse41.s
@@ -171,7 +171,7 @@ roundss $1, (%rax), %xmm2
# CHECK-NEXT: 3 2 1.00 * extractps $1, %xmm0, (%rax)
# CHECK-NEXT: 1 1 1.00 insertps $1, %xmm0, %xmm2
# CHECK-NEXT: 2 7 1.00 * insertps $1, (%rax), %xmm2
-# CHECK-NEXT: 1 6 0.50 * movntdqa (%rax), %xmm2
+# CHECK-NEXT: 2 7 0.50 * movntdqa (%rax), %xmm2
# CHECK-NEXT: 2 4 1.00 mpsadbw $1, %xmm0, %xmm2
# CHECK-NEXT: 3 10 1.00 * mpsadbw $1, (%rax), %xmm2
# CHECK-NEXT: 1 3 1.00 packusdw %xmm0, %xmm2
@@ -268,7 +268,7 @@ roundss $1, (%rax), %xmm2
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]
-# CHECK-NEXT: - - 36.67 46.17 22.00 22.00 2.50 49.17 - 2.50 2.50 2.50
+# CHECK-NEXT: - - 37.00 46.50 22.00 22.00 2.50 49.50 - 2.50 2.50 2.50
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions:
@@ -288,7 +288,7 @@ roundss $1, (%rax), %xmm2
# CHECK-NEXT: - - - - - - 0.50 1.00 - 0.50 0.50 0.50 extractps $1, %xmm0, (%rax)
# CHECK-NEXT: - - - - - - - 1.00 - - - - insertps $1, %xmm0, %xmm2
# CHECK-NEXT: - - - - 0.50 0.50 - 1.00 - - - - insertps $1, (%rax), %xmm2
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - - - movntdqa (%rax), %xmm2
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - - - movntdqa (%rax), %xmm2
# CHECK-NEXT: - - - 0.50 - - - 1.50 - - - - mpsadbw $1, %xmm0, %xmm2
# CHECK-NEXT: - - - 0.50 0.50 0.50 - 1.50 - - - - mpsadbw $1, (%rax), %xmm2
# CHECK-NEXT: - - - - - - - 1.00 - - - - packusdw %xmm0, %xmm2
diff --git a/llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s b/llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s
index e05911e4709dc1..1c08bb82d40061 100644
--- a/llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s
+++ b/llvm/test/tools/llvm-mca/X86/SkylakeClient/resources-avx1.s
@@ -1300,8 +1300,8 @@ vzeroupper
# CHECK-NEXT: 1 2 1.00 vmovmskps %ymm0, %ecx
# CHECK-NEXT: 2 1 1.00 * vmovntdq %xmm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * vmovntdq %ymm0, (%rax)
-# CHECK-NEXT: 1 6 0.50 * vmovntdqa (%rax), %xmm2
-# CHECK-NEXT: 1 7 0.50 * vmovntdqa (%rax), %ymm2
+# CHECK-NEXT: 2 7 0.50 * vmovntdqa (%rax), %xmm2
+# CHECK-NEXT: 2 8 0.50 * vmovntdqa (%rax), %ymm2
# CHECK-NEXT: 2 1 1.00 * vmovntpd %xmm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * vmovntpd %ymm0, (%rax)
# CHECK-NEXT: 2 1 1.00 * vmovntps %xmm0, (%rax)
@@ -1736,7 +1736,7 @@ vzeroupper
# CHECK: Resource pressure per iteration:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9]
-# CHECK-NEXT: - 126.00 338.58 200.58 173.83 173.83 38.00 324.58 6.25 11.33
+# CHECK-NEXT: - 126.00 339.25 201.25 173.83 173.83 38.00 325.25 6.25 11.33
# CHECK: Resource pressure by instruction:
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] Instructions:
@@ -2010,8 +2010,8 @@ vzeroupper
# CHECK-NEXT: - - 1.00 - - - - - - - vmovmskps %ymm0, %ecx
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 vmovntdq %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 vmovntdq %ymm0, (%rax)
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - vmovntdqa (%rax), %xmm2
-# CHECK-NEXT: - - - - 0.50 0.50 - - - - vmovntdqa (%rax), %ymm2
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - vmovntdqa (%rax), %xmm2
+# CHECK-NEXT: - - 0.33 0.33 0.50 0.50 - 0.33 - - vmovntdqa (%rax), %ymm2
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 vmovntpd %xmm0, (%rax)
# CHECK-NEXT: - - - - 0.33 0.33 1.00 - - 0.33 vmovntpd %ymm0, (%rax)
# CHECK-...
[truncated]
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ensure we use the SchedWriteVecMoveLSNT class for all (V)MOVNTDQA instructions, remove unnecessary scheduler overrides and adjust resource pipe usage to match uops.info/Agner numbers