Skip to content

JIT: Correct some AVX-512 instruction timings #126555

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
saucecontrol wants to merge 2 commits into from
Closed

JIT: Correct some AVX-512 instruction timings #126555

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
66ef08b
correct some AVX-512 instruction timings
saucecontrol Apr 5, 2026
f1738d4
feedback
saucecontrol Apr 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/coreclr/jit/emit.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2088,6 +2088,7 @@ class emitter
#define PERFSCORE_LATENCY_14C 14.0f
#define PERFSCORE_LATENCY_15C 15.0f
#define PERFSCORE_LATENCY_16C 16.0f
#define PERFSCORE_LATENCY_17C 17.0f
#define PERFSCORE_LATENCY_18C 18.0f
#define PERFSCORE_LATENCY_20C 20.0f
#define PERFSCORE_LATENCY_22C 22.0f
Expand Down
56 changes: 39 additions & 17 deletions src/coreclr/jit/emitxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20913,7 +20913,7 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
{
// ins reg, mem
result.insThroughput = PERFSCORE_THROUGHPUT_2X;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_3C : PERFSCORE_LATENCY_2C;
result.insLatency += opSize == EA_16BYTE ? PERFSCORE_LATENCY_2C : PERFSCORE_LATENCY_3C;
}
else
{
Expand Down Expand Up @@ -20943,11 +20943,12 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
break;
}

case INS_lddqu:
case INS_movntdqa:
{
assert(memAccessKind == PERFSCORE_MEMORY_READ);
result.insThroughput = PERFSCORE_THROUGHPUT_2X;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_3C : PERFSCORE_LATENCY_2C;
result.insLatency += opSize == EA_16BYTE ? PERFSCORE_LATENCY_2C : PERFSCORE_LATENCY_3C;
break;
}

Expand All @@ -20963,7 +20964,7 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
else if (memAccessKind == PERFSCORE_MEMORY_READ)
{
result.insThroughput = PERFSCORE_THROUGHPUT_2X;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_3C : PERFSCORE_LATENCY_2C;
result.insLatency += (EA_SIZE_IN_BYTES(opSize) <= 16) ? PERFSCORE_LATENCY_2C : PERFSCORE_LATENCY_3C;
}
else
{
Expand All @@ -20973,13 +20974,6 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
break;
}

case INS_lddqu:
{
result.insThroughput = PERFSCORE_THROUGHPUT_2X;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_3C : PERFSCORE_LATENCY_2C;
break;
}

case INS_vpmovdb:
case INS_vpmovdw:
case INS_vpmovqb:
Expand Down Expand Up @@ -21147,7 +21141,7 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
}
else
{
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_4C : PERFSCORE_LATENCY_2C;
result.insLatency += opSize == EA_16BYTE ? PERFSCORE_LATENCY_2C : PERFSCORE_LATENCY_4C;
result.insThroughput = PERFSCORE_THROUGHPUT_1C;
}
break;
Expand Down Expand Up @@ -21231,7 +21225,7 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
case INS_pmovzxdq:
{
result.insThroughput = PERFSCORE_THROUGHPUT_1C;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_3C : PERFSCORE_LATENCY_1C;
result.insLatency += opSize == EA_16BYTE ? PERFSCORE_LATENCY_1C : PERFSCORE_LATENCY_3C;
break;
}

Expand Down Expand Up @@ -21259,7 +21253,7 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
case INS_vcvtuqq2ps:
{
result.insThroughput = PERFSCORE_THROUGHPUT_1C;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_7C : PERFSCORE_LATENCY_5C;
result.insLatency += opSize == EA_16BYTE ? PERFSCORE_LATENCY_5C : PERFSCORE_LATENCY_7C;
break;
}

Expand Down Expand Up @@ -21349,8 +21343,22 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
case INS_vpgatherdd:
case INS_vgatherdps:
{
result.insThroughput = PERFSCORE_THROUGHPUT_4C;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_13C : PERFSCORE_LATENCY_11C;
if (opSize == EA_16BYTE)
{
result.insThroughput = PERFSCORE_THROUGHPUT_2C;
result.insLatency += PERFSCORE_LATENCY_11C;
}
else if (opSize == EA_32BYTE)
{
result.insThroughput = PERFSCORE_THROUGHPUT_4C;
result.insLatency += PERFSCORE_LATENCY_13C;
}
else
{
assert(opSize == EA_64BYTE);
result.insThroughput = PERFSCORE_THROUGHPUT_8C;
result.insLatency += PERFSCORE_LATENCY_17C;
}
break;
}

Expand All @@ -21361,8 +21369,22 @@ emitter::insExecutionCharacteristics emitter::getInsExecutionCharacteristics(ins
case INS_vgatherqps:
case INS_vgatherqpd:
{
result.insThroughput = PERFSCORE_THROUGHPUT_4C;
result.insLatency += opSize == EA_32BYTE ? PERFSCORE_LATENCY_11C : PERFSCORE_LATENCY_9C;
if (opSize == EA_16BYTE)
{
result.insThroughput = PERFSCORE_THROUGHPUT_1C;
result.insLatency += PERFSCORE_LATENCY_9C;
}
else if (opSize == EA_32BYTE)
{
result.insThroughput = PERFSCORE_THROUGHPUT_2C;
result.insLatency += PERFSCORE_LATENCY_11C;
}
else
{
assert(opSize == EA_64BYTE);
result.insThroughput = PERFSCORE_THROUGHPUT_4C;
result.insLatency += PERFSCORE_LATENCY_13C;
}
break;
}

Expand Down
Loading