zstd: extend executeSimple with history support #542
Merged
Conversation
This file contains 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
|
"Regressions" seems within margin of error. Good job, I will go through it. |
|
Impact of commit 3a35124 (comparison with the previous commit from this branch). Can't tell it's significantly better or worse, except few cases. |
|
So a tricky situation is that we cannot overread from history, since we may get out of the current page. So we need a "precise" memory copier. I have one for s2 which can be adapted. It is used for sizes 1->64, but it could be 0 -> 16 so it can be called whenever there is less than 16 bytes left. Untested, but this should be something like it: // func genMemMoveShort
// src and dst may not overlap.
// No registers are updated.
// Length must be 0 -> 16 bytes
func genMemMoveShort(name string, dst, src, length reg.GPVirtual, end LabelRef) {
Comment("genMemMoveShort")
AX, CX := GP64(), GP64()
name += "_memmove_"
// Only enable if length can be 0.
if true {
TESTQ(length, length)
JEQ(end)
}
CMPQ(length, U8(3))
JB(LabelRef(name + "move_1or2"))
JE(LabelRef(name + "move_3"))
CMPQ(length, U8(8))
JB(LabelRef(name + "move_4through7"))
//Label(name + "move_8through16")
MOVQ(Mem{Base: src}, AX)
MOVQ(Mem{Base: src, Disp: -8, Index: length, Scale: 1}, CX)
MOVQ(AX, Mem{Base: dst})
MOVQ(CX, Mem{Base: dst, Disp: -8, Index: length, Scale: 1})
JMP(end)
Label(name + "move_1or2")
MOVB(Mem{Base: src}, AX.As8())
MOVB(Mem{Base: src, Disp: -1, Index: length, Scale: 1}, CX.As8())
MOVB(AX.As8(), Mem{Base: dst})
MOVB(CX.As8(), Mem{Base: dst, Disp: -1, Index: length, Scale: 1})
JMP(end)
Label(name + "move_3")
MOVW(Mem{Base: src}, AX.As16())
MOVB(Mem{Base: src, Disp: 2}, CX.As8())
MOVW(AX.As16(), Mem{Base: dst})
MOVB(CX.As8(), Mem{Base: dst, Disp: 2})
JMP(end)
Label(name + "move_4through7")
MOVL(Mem{Base: src}, AX.As32())
MOVL(Mem{Base: src, Disp: -4, Index: length, Scale: 1}, CX.As32())
MOVL(AX.As32(), Mem{Base: dst})
MOVL(CX.As32(), Mem{Base: dst, Disp: -4, Index: length, Scale: 1})
JMP(end)
} |
|
Right, I will change it. |
WojciechMula
force-pushed
the
asm-seqdec-execute-history
branch
from
3a35124 to
83ad8f9
Compare
|
OK, I used another memory copy routine for history. Updated the main issue with current timings. Still nice improvements for cases with history. |
klauspost
requested changes
Apr 4, 2022
Co-authored-by: Klaus Post <klauspost@gmail.com>
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.
Add history support in the asm implementation. Part of task #515.
As usual marking as a draft, because of few failing tests.[fixed]Performance comparison between the current master and this branch on an IceLake machine with the hacked
decodeSyncis below. There are some nice speedups, but there are also regressions for almost all cases without history. To overcome that, we can have two specialisations:executeWithoutDictionaryandexecuteWithoutDictionatyAndHistory- what do you think?