Bug 146484

Summary:

jsc-tailcall: Add new opcodes for tail calls

Product:

WebKit

Reporter:

Basile Clement <basile_clement>

Component:

JavaScriptCore

Assignee:

Basile Clement <basile_clement>

Status:

RESOLVED FIXED

Severity:

Enhancement

CC:

fpizlo, ggaren, mark.lam, msaboff, saam

Priority:

Version:

528+ (Nightly build)

Hardware:

All

OS:

All

Bug Depends on:

146481, 146846, 146847, 146967

Bug Blocks:

146477, 146850, 147195

Attachments:

Description	Flags
Patch	msaboff: review+
ARM64 test output with failures	none
ARM32 test output with failures	none

Basile Clement

Reported 2015-06-30 17:43:44 PDT

For now, this opcode should be treated as a regular call in all backends.

Attachments
Patch (117.19 KB, patch) 2015-07-21 15:49 PDT, Basile Clement	msaboff: review+	Details Formatted Diff Diff
ARM64 test output with failures (deleted) 2015-07-22 13:21 PDT, Michael Saboff	no flags	Details
ARM32 test output with failures (16.93 MB, application/octet-stream) 2015-07-22 15:56 PDT, Michael Saboff	no flags	Details
View All Add attachment proposed patch, testcase, etc.

Basile Clement

Comment 1 2015-07-10 12:35:46 PDT

(In reply to comment #0) > For now, this opcode should be treated as a regular call in all backends.

Basile Clement

Comment 2 2015-07-10 12:36:13 PDT

(In reply to comment #1) > (In reply to comment #0) > > For now, this opcode should be treated as a regular call in all backends. (Oops, messed this up) We should not do this, since it is a semantics change. Instead, we should implement this in the LLint and baseline JIT first, then progressively implement it in the DFG and FTL.

Basile Clement

Comment 3 2015-07-21 15:49:06 PDT

Created attachment 257212 [details] Patch

Michael Saboff

Comment 4 2015-07-22 10:18:05 PDT

Comment on attachment 257212 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=257212&action=review r=me with minor fix > Source/JavaScriptCore/llint/LowLevelInterpreter.asm:609 > + addp CallerFrameAndPCSize, sp I don't think ARMv7 will allow you to do math with sp register. You'll probably need to use a temp and then move result to sp.

Mark Lam

Comment 5 2015-07-22 10:19:36 PDT

(In reply to comment #4) > > Source/JavaScriptCore/llint/LowLevelInterpreter.asm:609 > > + addp CallerFrameAndPCSize, sp > > I don't think ARMv7 will allow you to do math with sp register. You'll > probably need to use a temp and then move result to sp. Are you sure? I thought ARMv7 does allow it (hardware-wise). Should just try a build and see what gets generated.

Basile Clement

Comment 6 2015-07-22 10:46:07 PDT

(In reply to comment #5) > (In reply to comment #4) > > > Source/JavaScriptCore/llint/LowLevelInterpreter.asm:609 > > > + addp CallerFrameAndPCSize, sp > > > > I don't think ARMv7 will allow you to do math with sp register. You'll > > probably need to use a temp and then move result to sp. > > Are you sure? I thought ARMv7 does allow it (hardware-wise). Should just > try a build and see what gets generated. We established it was fine since we are using the two-argument version. Old version was trying to use the three-mode operand by adding to calleeFrame, which wouldn't work. Landing this soon.

Basile Clement

Comment 7 2015-07-22 10:58:06 PDT

Landed in r187166 <http://trac.webkit.org/changeset/187166>.

Michael Saboff

Comment 8 2015-07-22 13:21:28 PDT

Created attachment 257287 [details] ARM64 test output with failures

Michael Saboff

Comment 9 2015-07-22 15:56:56 PDT

Created attachment 257306 [details] ARM32 test output with failures

Note You need to log in before you can comment on or make changes to this bug.