Closed gwoltman closed 7 years ago
and vmovdqa32 ZMMWORD PTR [rdi][rdx], zmm14 assembles as vmovdqa32 %zmm30,(%rdx,%r15,1)
Working on them now, one more change to make to get them all fixed and we’ll have another update.
From: gwoltman [mailto:notifications@github.com] Sent: 26 November 2016 09:22 PM To: Terraspace/HJWasm HJWasm@noreply.github.com Subject: Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
and vmovdqa32 ZMMWORD PTR [rdi][rdx], zmm14 assembles as vmovdqa32 %zmm30,(%rdx,%r15,1)
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263086683 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVFeHBxxzad3ubcIfZdChxOfl9kRiks5rCKMVgaJpZM4K8_lF .
Hi,
Update (dated 29th) is available for download, it should fix this latest set of issues.
Regards,
John
From: gwoltman [mailto:notifications@github.com] Sent: 26 November 2016 09:22 PM To: Terraspace/HJWasm HJWasm@noreply.github.com Subject: Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
and vmovdqa32 ZMMWORD PTR [rdi][rdx], zmm14 assembles as vmovdqa32 %zmm30,(%rdx,%r15,1)
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263086683 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVFeHBxxzad3ubcIfZdChxOfl9kRiks5rCKMVgaJpZM4K8_lF .
Next: vpsrad zmm15, zmm0, 31 assembles as: vpsrad $0x1f,0x0(%rax,%rax,8),%zmm15
And: vpsrad ymm15, ymm8, 31 assembles as accoring to gdb: (bad) but executes OK.
Fixing now..
From: gwoltman [mailto:notifications@github.com] Sent: 29 November 2016 08:27 PM To: Terraspace/HJWasm HJWasm@noreply.github.com Cc: John Hankinson john@terraspace.co.uk; Comment comment@noreply.github.com Subject: {Spam?} Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
Next: vpsrad zmm15, zmm0, 31 assembles as: vpsrad $0x1f,0x0(%rax,%rax,8),%zmm15
And: vpsrad ymm15, ymm8, 31 assembles as accoring to gdb: (bad) but executes OK.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263688487 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVG38kw1Se1AgyO63Gjsebi-hGASTks5rDIqHgaJpZM4K8_lF .
Fixes up, please test again.
From: gwoltman [mailto:notifications@github.com] Sent: 29 November 2016 08:27 PM To: Terraspace/HJWasm HJWasm@noreply.github.com Cc: John Hankinson john@terraspace.co.uk; Comment comment@noreply.github.com Subject: {Spam?} Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
Next: vpsrad zmm15, zmm0, 31 assembles as: vpsrad $0x1f,0x0(%rax,%rax,8),%zmm15
And: vpsrad ymm15, ymm8, 31 assembles as accoring to gdb: (bad) but executes OK.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263688487 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVG38kw1Se1AgyO63Gjsebi-hGASTks5rDIqHgaJpZM4K8_lF .
Those 2 are fixed.
More cases where gdb does not disassemble properly: vpsrlq ymm9, ymm9, 30 dumps as: c4 c1 35 73 (bad)
I'm debugging a different problem now. I may have more to report later.
Just tried this one with the latest updated package:
I’m getting:
000000013fd41000 C4 C1 35 73 D1 1E vpsrlq ymm9, ymm9, 0x1e
Which is correct and reported the same via Defuse:
Assembly
Raw Hex (zero bytes in bold):
C4C13573D11E
String Literal:
"\xC4\xC1\x35\x73\xD1\x1E"
Array Literal:
{ 0xC4, 0xC1, 0x35, 0x73, 0xD1, 0x1E }
Disassembly:
0: c4 c1 35 73 d1 1e vpsrlq ymm9,ymm9,0x1e
From: gwoltman [mailto:notifications@github.com] Sent: 29 November 2016 10:47 PM To: Terraspace/HJWasm HJWasm@noreply.github.com Cc: John Hankinson john@terraspace.co.uk; Comment comment@noreply.github.com Subject: Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
Those 2 are fixed.
More cases where gdb does not disassemble properly: vpsrlq ymm9, ymm9, 30 dumps as: c4 c1 35 73 (bad)
I'm debugging a different problem now. I may have more to report later.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263725300 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVGzu2P6l_Xwu-iUbW0AHmRvtw6mZks5rDKt2gaJpZM4K8_lF .
Hmm, you are right. I must have objdump'ed an older object file.
How about: vaddpd zmm2, zmm2, zmm3 Do you get this? 62 d1 ed 48 58 d3 vaddpd %zmm11,%zmm2,%zmm2
and: vcmpeqpd k1, zmm2, zmm0 vmovapd zmm5, zmm6 vmulpd zmm2, zmm28, zmm6 assembles as: 62 d1 ed 48 c2 c8 00 vcmpeqpd %zmm8,%zmm2,%k1 62 d1 fd 48 28 ee vmovapd %zmm14,%zmm5 62 d1 9d 40 59 d6 vmulpd %zmm14,%zmm28,%zmm2
New build up today 30th with fixes.
From: gwoltman [mailto:notifications@github.com] Sent: 30 November 2016 01:58 AM To: Terraspace/HJWasm HJWasm@noreply.github.com Cc: John Hankinson john@terraspace.co.uk; Comment comment@noreply.github.com Subject: Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
and: vcmpeqpd k1, zmm2, zmm0 vmovapd zmm5, zmm6 vmulpd zmm2, zmm28, zmm6 assembles as: 62 d1 ed 48 c2 c8 00 vcmpeqpd %zmm8,%zmm2,%k1 62 d1 fd 48 28 ee vmovapd %zmm14,%zmm5 62 d1 9d 40 59 d6 vmulpd %zmm14,%zmm28,%zmm2
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-263759973 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVHtIPuJ6h-nDn76j4Oc0UZOw3X8Rks5rDNgPgaJpZM4K8_lF .
those are done, next up:
vdivpd zmm15, zmm27, zmm17
vaddpd zmm16, zmm16, zmm17
vmulpd zmm17, zmm28, zmm17
vmovapd zmm19, zmm20
assemble as (according to objconv):
vdivpd zmm15, zmm27, zmm25 ; 7051 62 11 A5 40: 5E. F9
vaddpd zmm16, zmm16, zmm25 ; 7A31 62 81 FD 40: 58. C1
vmulpd zmm17, zmm28, zmm25 ; 7337 62 81 9D 40: 59. C9
vmovapd zmm19, zmm28 ; 7918 62 81 FD 48: 28. DC
These are all fixed now.
One note however:
vcvtpd2ps ymm0,xmm1
vcvtpd2ps ymm0,xmmword ptr mop
;000000013f6c1274 C5 FD 5A C1 vcvtpd2ps xmm0, ymm1
;000000013f6c1278 C5 FD 5A 05 80 2D 00 00 vcvtpd2ps xmm0, ymmword ptr [rip+0x2d80]
This has been adjusted to now produce an error during assembly, as the operands should be xmm then ymm, (so they’re back to front).
From: gwoltman [mailto:notifications@github.com] Sent: 01 December 2016 12:48 AM To: Terraspace/HJWasm HJWasm@noreply.github.com Cc: John Hankinson john@terraspace.co.uk; Comment comment@noreply.github.com Subject: Re: [Terraspace/HJWasm] Back to AVX512 development (#43)
those are done, next up: vdivpd zmm15, zmm27, zmm17 vaddpd zmm16, zmm16, zmm17 vmulpd zmm17, zmm28, zmm17 vmovapd zmm19, zmm20 assemble as (according to objconv): vdivpd zmm15, zmm27, zmm25 ; 7051 62 11 A5 40: 5E. F9 vaddpd zmm16, zmm16, zmm25 ; 7A31 62 81 FD 40: 58. C1 vmulpd zmm17, zmm28, zmm25 ; 7337 62 81 9D 40: 59. C9 vmovapd zmm19, zmm28 ; 7918 62 81 FD 48: 28. DC
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/Terraspace/HJWasm/issues/43#issuecomment-264044930 , or mute the thread https://github.com/notifications/unsubscribe-auth/AQGQVLGa6V10oGNLS7jss51Yk3xK6EQPks5rDhlWgaJpZM4K8_lF .
No issues at present! Well done.
assembles as vmovdqu32 (%rdx,%r14,1),%zmm14 according to gcc ... that looks right to me.
But, vmovdqa ZMMWORD PTR [rdi][rdx], zmm14 assembles as: vmovdqa32 %xmm14,(%rdx,%r15,1) according to gcc.
Also, this vpandn ymm14, ymm15, ymm14 assembles as: (bad) according to gcc. But it does execute properly.