From: Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Maxim Kokryashkin <m.kokryashkin@tarantool.org>,
Sergey Bronnikov <sergeyb@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org
Subject: [Tarantool-patches] [PATCH luajit] Fix jit.dump() output for IR_CONV.
Date: Wed, 25 Oct 2023 16:14:28 +0300 [thread overview]
Message-ID: <20231025131428.4124-1-skaplun@tarantool.org> (raw)
From: Mike Pall <mike>
(cherry-picked from commit 44bd7437a27e0d19bcf878c20ad27a673f17f40b)
When dumping IRs via `jit.dump()`, the incorrect value of `IR_CONV` mode
shift (`IRCONV_CSH`) is used. This patch fixes the value.
Sergey Kaplun:
* added the description and the test for the problem
Part of tarantool/tarantool#9145
---
Branch: https://github.com/tarantool/luajit/tree/skaplun/fix-ir-conv
Tarantool PR: https://github.com/tarantool/tarantool/pull/9307
Issue: https://github.com/tarantool/tarantool/issues/9145
src/jit/dump.lua | 2 +-
.../fix-jit-dump-ir-conv.test.lua | 68 +++++++++++++++++++
2 files changed, 69 insertions(+), 1 deletion(-)
create mode 100644 test/tarantool-tests/fix-jit-dump-ir-conv.test.lua
diff --git a/src/jit/dump.lua b/src/jit/dump.lua
index 27b5c2ae..84fe0044 100644
--- a/src/jit/dump.lua
+++ b/src/jit/dump.lua
@@ -277,7 +277,7 @@ local litname = {
local s = irtype[band(mode, 31)]
s = irtype[band(shr(mode, 5), 31)].."."..s
if band(mode, 0x800) ~= 0 then s = s.." sext" end
- local c = shr(mode, 14)
+ local c = shr(mode, 12)
if c == 2 then s = s.." index" elseif c == 3 then s = s.." check" end
t[mode] = s
return s
diff --git a/test/tarantool-tests/fix-jit-dump-ir-conv.test.lua b/test/tarantool-tests/fix-jit-dump-ir-conv.test.lua
new file mode 100644
index 00000000..ed03210a
--- /dev/null
+++ b/test/tarantool-tests/fix-jit-dump-ir-conv.test.lua
@@ -0,0 +1,68 @@
+local tap = require('tap')
+local test = tap.test('fix-jit-dump-ir-conv'):skipcond({
+ ['Test requires JIT enabled'] = not jit.status(),
+ ['Disabled on *BSD due to #4819'] = jit.os == 'BSD',
+})
+
+test:plan(2)
+
+-- Test file to demonstrate LuaJIT incorrect `jit.dump()` output
+-- for `IR_CONV`.
+
+local jparse = require('utils').jit.parse
+
+-- XXX: Avoid any traces compilation due to hotcount collisions
+-- for predictable results.
+jit.off()
+jit.flush()
+
+jit.opt.start('hotloop=1')
+
+jit.on()
+jparse.start('i')
+
+-- luacheck: ignore
+local tab = {}
+local idx = 1
+for _ = 1, 4 do
+ -- `int IR_CONV int.num index`.
+ tab[idx] = idx
+end
+
+local traces = jparse.finish()
+
+-- Skip tests for DUALNUM mode since it has no conversions (for
+-- the same cases).
+local IS_DUALNUM = not traces[1]:has_ir('num SLOAD')
+
+test:ok(IS_DUALNUM or traces[1]:has_ir('CONV.*int.num index'),
+ 'correct dump for index')
+
+local function trace(step)
+ -- `int IR_CONV int.num check` for `step` conversion.
+ for _ = 1, 4, step do
+ end
+end
+
+-- XXX: Reset hotcounters and traces.
+jit.flush()
+jit.opt.start('hotloop=1')
+
+jparse.start('i')
+-- Compile the inner trace first.
+trace(1)
+-- Compile the big trace with the first trace inlined.
+-- Needs narrowing optimization enabled.
+-- XXX: Reset hotcounters and call the function 3 times to
+-- compile. Needed to avoid hotcount collisions.
+jit.opt.start('hotloop=1')
+trace(1)
+trace(1)
+trace(1)
+
+traces = jparse.finish()
+
+test:ok(IS_DUALNUM or traces[2]:has_ir('CONV.*int.num check'),
+ 'correct dump for check')
+
+test:done(true)
--
2.42.0
next reply other threads:[~2023-10-25 13:19 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-10-25 13:14 Sergey Kaplun via Tarantool-patches [this message]
2023-10-26 12:24 ` Maxim Kokryashkin via Tarantool-patches
2023-11-08 19:29 ` Sergey Bronnikov via Tarantool-patches
2023-11-23 6:32 ` Igor Munkin via Tarantool-patches
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20231025131428.4124-1-skaplun@tarantool.org \
--to=tarantool-patches@dev.tarantool.org \
--cc=m.kokryashkin@tarantool.org \
--cc=sergeyb@tarantool.org \
--cc=skaplun@tarantool.org \
--subject='Re: [Tarantool-patches] [PATCH luajit] Fix jit.dump() output for IR_CONV.' \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox