Hi, Stefan, What exact error does it print?

Because I just find the assertion failure at lj_state.c:174:

luajit: lj_state.c:174: close_state: Assertion `g->gc.total == sizeof(GG_State)' failed.

It seems my add interleaved IR dump patch (commit id: 4e6f1de6d6befdf729f0fc2eefb96099a7eddaf3 ) introduces this error. I'll take a look at it.

-- Best regards, Zhongwei

On 14 July 2016 at 17:20, Stefan Pejic stefan.pejic@rt-rk.com wrote:

Yes, that assertion is not good, but I actually think it should be completely removed, because ARM64 is always dual-number and it would never fail.

Anyway, that's not where I noticed an issue, I was getting a failure here: lj_state.c:172. This is the example I was running:

do -- IR_HLOAD: nil & pri local m = { x = nil, y = true } for i=1,100 do if not m.x then m.y = not m.y end end assert(m.y) end

On Thu, 14 Jul 2016 09:29:53 +0100 Charles Baylis charles.baylis@linaro.org wrote:

...

On 13 July 2016 at 14:07, Stefan Pejic stefan.pejic@rt-rk.com wrote:

...

Hello,

I ran into an issue with asm_intcomp. Is anyone working on this at the moment?

I'm not looking at that.

...

Also, I noticed that enabling assertions in src/Makefile causes some tests failures. And that's happening only after the latest merges.

I think the assert in asm_sload: lua_assert(!irt_isint(t) || (ir->op2 & (IRSLOAD_CONVERT|IRSLOAD_FRAME))); should be lua_assert(LJ_DUALNUM || !irt_isint(t) || (ir->op2 & (IRSLOAD_CONVERT|IRSLOAD_FRAME)));

which would match the definition in lj_asm_x86.h. Does that fix the problem you see?

Charles

---

luajit mailing list luajit@lists.linaro.org https://lists.linaro.org/mailman/listinfo/luajit

Sorry for late reply. This issue is fixed now and patch is merged.

-- Best regards, Zhongwei

On 14 July 2016 at 18:17, Stefan Pejic stefan.pejic@rt-rk.com wrote:

Hey,

That is exactly what I'm getting as well.

Stefan

On Thu, 14 Jul 2016 18:02:00 +0800 Zhongwei Yao zhongwei.yao@linaro.org wrote:

...

Hi, Stefan, What exact error does it print?

Because I just find the assertion failure at lj_state.c:174:

luajit: lj_state.c:174: close_state: Assertion `g->gc.total == sizeof(GG_State)' failed.

It seems my add interleaved IR dump patch (commit id: 4e6f1de6d6befdf729f0fc2eefb96099a7eddaf3 ) introduces this error. I'll take a look at it.

-- Best regards, Zhongwei

On 14 July 2016 at 17:20, Stefan Pejic stefan.pejic@rt-rk.com wrote:

...

Yes, that assertion is not good, but I actually think it should be completely removed, because ARM64 is always dual-number and it would never fail.

Anyway, that's not where I noticed an issue, I was getting a failure here: lj_state.c:172. This is the example I was running:

do -- IR_HLOAD: nil & pri local m = { x = nil, y = true } for i=1,100 do if not m.x then m.y = not m.y end end assert(m.y) end

On Thu, 14 Jul 2016 09:29:53 +0100 Charles Baylis charles.baylis@linaro.org wrote:

...

On 13 July 2016 at 14:07, Stefan Pejic stefan.pejic@rt-rk.com wrote:

...

Hello,

I ran into an issue with asm_intcomp. Is anyone working on this at the moment?

I'm not looking at that.

...

Also, I noticed that enabling assertions in src/Makefile causes some tests failures. And that's happening only after the latest merges.

I think the assert in asm_sload: lua_assert(!irt_isint(t) || (ir->op2 & (IRSLOAD_CONVERT|IRSLOAD_FRAME))); should be lua_assert(LJ_DUALNUM || !irt_isint(t) || (ir->op2 & (IRSLOAD_CONVERT|IRSLOAD_FRAME)));

which would match the definition in lj_asm_x86.h. Does that fix the problem you see?

Charles

---

luajit mailing list luajit@lists.linaro.org https://lists.linaro.org/mailman/listinfo/luajit