847220 - nanojit-lirasm need updated

Status: RESOLVED INVALID

(Core Graveyard :: Nanojit, defect)

Description

[Domingo Alvarez Duarte]

User Agent: Mozilla/5.0 (X11; Ubuntu; Linux i686; rv:19.0) Gecko/20100101 Firefox/19.0
Build ID: 20130227155321

Steps to reproduce:

I tried to compile the  nanojit-lirasm and it doesn't compile, there is dependency not solved for flot4Support, when manually including float4Support.(h & cpp), there is a dependency not solved with "sqrtf", commenting all code that uses "sqrtf", there is a problem with "Lirasm" class constructor complaining about missong Codealloc() constructor without arguments:

Lirasm::Lirasm(bool verbose, Config& config) :
    mConfig(config),mCodeAlloc(&mConfig), ////// adding mCodeAlloc(&mConfig), solve it
    mAssm(mCodeAlloc, mAlloc, mAlloc, &mLogc, mConfig)


Actual results:

Several errors reported by the compiler related to the problems reported.
After the process described above, I run "make check" and get this failures:
...
TEST-PASS | lirasm | lirasm  --execute --random 1000000
Assertion failure: memcmp( &f4, &other, sizeof(float4_t)) == 0 (./nanojit/LIR.cpp:3401)
TEST-UNEXPECTED-FAIL | lirasm | lirasm  --execute --random 1000000 --optimize
expected output
Output is: 0
actual output
TEST-PASS | lirasm | lirasm --nosse --execute ./lirasm/tests/./addd.in
...
Assertion failure: _config.i386_sse2 (./nanojit/Nativei386.cpp:1918)
TEST-UNEXPECTED-FAIL | lirasm | lirasm --nosse --execute --random 1000000
expected output
Output is: 0
actual output
make: *** [check] Error 1



Expected results:

Compile and produce "lirasm" executable and also pass all tests.

Comment 1

[Domingo Alvarez Duarte]

/* config.h.  Generated from config.h.in by configure.  */
/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */


#define DEBUG 1
/* #undef AVMPLUS_AMD64 */
/* #undef AVMPLUS_64BIT */
/* #undef AVMPLUS_ARM */
#define AVMPLUS_IA32 1
#define AVMPLUS_LINUX 1
/* #undef AVMPLUS_OS2 */
/* #undef AVMPLUS_PPC */
/* #undef AVMPLUS_MIPS */
/* #undef VMCFG_SH4 */
/* #undef VMCFG_FLOAT */
#define VMCFG_SSE2 1
/* #undef VMCFG_NEON */
/* #undef AVMPLUS_SPARC */
/* #undef SOLARIS */
#define AVMPLUS_UNIX 1
/* #undef AVMPLUS_WIN32 */
#define FEATURE_NANOJIT 1
/* #undef WINVER */
/* #undef _CRT_NONSTDC_NO_DEPRECATE */
/* #undef _CRT_NONSTDC_NO_WARNINGS */
/* #undef _CRT_SECURE_NO_DEPRECATE */
/* #undef _CRT_SECURE_NO_WARNINGS */
/* #undef WIN32 */
/* #undef _WIN32_IE */
/* #undef _WIN32_WINNT */

Comment 2

[Domingo Alvarez Duarte]

Instead of commenting out references to "sqrtf" now I added a define prior to it's usage on float4Support and could compile it as well but with the same errors when "make check":

#define sqrtf sqrt
// in hardware: vdupq_n_f32, or _mm_sqrt_ps
REALLY_INLINE float4_t f4_sqrt(const float4_t& v) 
{ 
    float4_t retval = { sqrtf(v.x), sqrtf(v.y), sqrtf(v.z), sqrtf(v.w)};
    return retval;
}

Comment 3

[Domingo Alvarez Duarte]

I did some modifications to see why the test fails:
        //NanoAssert(VMPI_memcmp( &f4, &other, sizeof(float4_t)) == 0);
	if(!f4_eq_i(f4, other)){
		printf("%d :: %d :: %.17f:%.17f, %.17f:%.17f, %.17f:%.17f, %.17f:%.17f\n", 
			f4_eq_i(f4, other),  VMPI_memcmp( &f4, &other, sizeof(float4_t)),
			f4.x, other.x, f4.y, other.y, f4.w, other.w, f4.z, other.z);
		//printf("%d  %d\n", f4_eq_i(f4, other), VMPI_memcmp( &f4, &other, sizeof(float4_t)));
	}
	//NanoAssert(f4_eq_i(f4, other));

And I've got this output on one run:

TEST-UNEXPECTED-FAIL | lirasm | lirasm  --execute --random 1000000 --optimize
expected output
Output is: 0
actual output
0 :: -1 :: nan:nan, 180.00000000000000000:180.00000000000000000, 0.00000000000000000:0.00000000000000000, 0.00000000000039189:0.00000000000039189
0 :: 0 :: -nan:-nan, 0.00000000000000000:0.00000000000000000, 2.40486621856689453:2.40486621856689453, 0.00000000000000000:0.00000000000000000
0 :: 0 :: -nan:-nan, 0.00000000000000000:0.00000000000000000, 0.00000000000000000:0.00000000000000000, 0.00000000000000000:0.00000000000000000

I fails due to the presence of "nan" or "-nan" and obviously on that situation the other bits contains rubish.

The "nan" was expected to happen or it is a bug ?

Comment 4

[Benjamin Bouvier [:bbouvier] (inactive)]

Nanojit is long gone.