Jerome Kieffer comments

Results 408 comments of


                                            Jerome Kieffer

Problem converting CBF files to SFRM

Maybe, I did not find anything like this in Bruker's documentation. In version 0.11 I added a few header which were missing like "VERSION" and "FORMAT", now `adxv` recognizes some...

Consolidate SSE optimized shuffle with Blosc.

To emphasize on this, the bitunshuffle uses an extra temporary buffer betweed the byte-transpose and the bitwise transpose. I believe this reduces drastically the performances when the buffer size increases...

Consolidate SSE optimized shuffle with Blosc.

On Fri, 08 Nov 2019 05:55:13 -0800 Kiyoshi Masui wrote: > Keep in mind that the whole operation is blocked in small blocks of 8k bytes, which does fit in...

Consolidate SSE optimized shuffle with Blosc.

> Looking at your slides, you are testing the version of bitshuffle that ships with blosc? yes, blosc2 beta to be precise. > Can you post some test code? For...

Numerical precision issue

In my tests, what fails is multiplication and division. Thus, I am running the conformance tests on add, subtract, multiply and divide to validate if they are properly executed...

Numerical precision issue

Apparently all those tests passed, but the division is apparently performed with reduced precision: ``` lintaillefer:build/test_conformance/math_brute_force % ./test_bruteforce CL_DEVICE_TYPE_GPU -d -m 93:101 |tee log_4op.txt ./test_bruteforce CL_DEVICE_TYPE_GPU -d -m 93:101sizeof( void*)...

Numerical precision issue

For a reminder, here are the failing tests (with errors): ``` TestOptions: WITH_QT_TEST=True WITH_OPENCL_TEST=True WITH_GL_TEST=True TEST_LOW_MEM=False TEST_IS32_BIT=False TEST_RANDOM=False F...FF... ====================================================================== FAIL: test_dw_div_dw (pyFAI.opencl.test.test_openCL.TestDoubleWord.test_dw_div_dw) ---------------------------------------------------------------------- Traceback (most recent call last): File...