[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #21 from Ivo Raisr --- Thank you all for your responses! They were really helpful. -- You are receiving this mail because: You are watching all bug changes.

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution|---

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #19 from Ivo Raisr --- (In reply to Julian Seward from comment #17) > I expect arm(32) would also get a quite big improvement, because > it's also pretty low on registers. [This is not a request to measure, > just random

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #18 from aleksandra --- I tested this new patches and here are results: MIPS64 both transtab 1,837 (44,740 -> 868,632; ratio 19.4) [0 scs] avg tce size 472 neither transtab 1,834 (44,664 -> 865,872;

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #17 from Julian Seward --- (In reply to Ivo Raisr from comment #16) > Here are the performance numbers for amd64 and x86: That's some nice improvements, especially for x86. Thanks for the measurements. I expect arm(32)

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #16 from Ivo Raisr --- I've fixed the problem with duplicate hregIsVirtual() in host_arch_defs.c files. Here are the performance numbers for amd64 and x86: Running inner Memcheck on perf/bz2 (compiled with -O or -O2).

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Attachment #108135|0 |1 is obsolete|

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Attachment #108134|0 |1 is obsolete|

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #13 from Julian Seward --- (In reply to aleksandra from comment #12) > Created attachment 108160 [details] > mips report > I hope this is better. Thanks, yes. So, what we can pull from that is MIPS64 bothtranstab:

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #12 from aleksandra --- Created attachment 108160 --> https://bugs.kde.org/attachment.cgi?id=108160=edit mips report I hope this is better. -- You are receiving this mail because: You are watching all bug

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #11 from Julian Seward --- (In reply to aleksandra from comment #10) > Created attachment 108138 [details] > mips report Aleksandra, thank you for testing on MIPS. Please can you re-attach the report as a plain text file?

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 aleksandra changed: What|Removed |Added CC|

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #9 from Julian Seward --- (In reply to Ivo Raisr from comment #6) > amd64: > vanilla:45,112,349,784 total; 165,978,807 reg alloc; ratio 15.5 > v3-reoder: 44,943,765,809 total; 167,403,237 reg alloc; ratio 15.3 Nice.

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #8 from Julian Seward --- (In reply to Ivo Raisr from comment #5) > Created attachment 108135 [details] > register allocation Looks plausible to me, although I can't make any meaningful assessment of it. -- You are

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Julian Seward changed: What|Removed |Added CC||jsew...@acm.org --- Comment

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #6 from Ivo Raisr --- I've tested on amd64, ppc8le and arm64 architectures. In all cases, the produced code was more compact and overall performance better, when running inner Memcheck on perf/bz2. Numbers are given as

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Attachment #107964|0 |1 is obsolete|

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Attachment #107963|0 |1 is obsolete|

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added See Also||https://bugs.kde.org/show_b

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #3 from Ivo Raisr --- The only remaining thing is to refactor isMove callbacks in the remaining architectures so that the information is available directly in HRegUsage. -- You are receiving this mail because: You are

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 --- Comment #2 from Ivo Raisr --- Created attachment 107964 --> https://bugs.kde.org/attachment.cgi?id=107964=edit register allocation -- You are receiving this mail because: You are watching all bug changes.

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Status|CONFIRMED |ASSIGNED -- You are receiving

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added CC||iv...@ivosh.net --- Comment #1

[valgrind] [Bug 384987] VEX register allocator: allocate caller-save registers for short lived vregs

https://bugs.kde.org/show_bug.cgi?id=384987 Ivo Raisr changed: What|Removed |Added Ever confirmed|0 |1