Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Use AVX512 to zero locals #91166

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

Merged
merged 2 commits into from
Aug 28, 2023
Merged

Use AVX512 to zero locals #91166

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
61a05ee
Use AVX512 to zero locals
EgorBo Aug 27, 2023
8675483
Fix loop
EgorBo Aug 27, 2023
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
19 changes: 17 additions & 2 deletions src/coreclr/jit/codegenxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10894,9 +10894,12 @@ void CodeGen::genZeroInitFrameUsingBlockInit(int untrLclHi, int untrLclLo, regNu
assert((blkSize + alignmentHiBlkSize) == (untrLclHi - untrLclLo));
#endif // !defined(TARGET_AMD64)

const int maxSimdSize = (int)compiler->roundDownSIMDSize(blkSize);
assert((maxSimdSize >= XMM_REGSIZE_BYTES) && (maxSimdSize <= ZMM_REGSIZE_BYTES));

// The loop is unrolled 3 times so we do not move to the loop block until it
// will loop at least once so the threshold is 6.
if (blkSize < (6 * XMM_REGSIZE_BYTES))
if (blkSize < (6 * maxSimdSize))
{
// Generate the following code:
//
Expand All @@ -10905,10 +10908,22 @@ void CodeGen::genZeroInitFrameUsingBlockInit(int untrLclHi, int untrLclLo, regNu
// ...
// movups xmmword ptr [ebp/esp-OFFS], xmm4
// mov qword ptr [ebp/esp-OFFS], rax

//
// NOTE: it implicitly zeroes YMM4 and ZMM4 as well.
emit->emitIns_SIMD_R_R_R(INS_xorps, EA_16BYTE, zeroSIMDReg, zeroSIMDReg, zeroSIMDReg);

int i = 0;
if (maxSimdSize > XMM_REGSIZE_BYTES)
{
for (; i <= blkSize - maxSimdSize; i += maxSimdSize)
{
// We previously aligned data to 16 bytes which might not be aligned to maxSimdSize
emit->emitIns_AR_R(simdUnalignedMovIns(), EA_ATTR(maxSimdSize), zeroSIMDReg, frameReg,
alignedLclLo + i);
}
// Remainder will be handled by the xmm loop below
}

for (; i < blkSize; i += XMM_REGSIZE_BYTES)
{
emit->emitIns_AR_R(simdMov, EA_ATTR(XMM_REGSIZE_BYTES), zeroSIMDReg, frameReg, alignedLclLo + i);
Expand Down