Old AMD GPU does not successfully compile OpenCL C code about fluidx3d

[Yujif1Aero]

Thank you for replying @ProjectPhysX . Sorry, your modified code doesnt work.
However, I have tried to modify your code, and I have got the solution.
I have modified TYPE number like this,

Due to it, your code is working now!!!

Could you review this? Let me know this solution is correct or not,
**I use as same kernel.cpp as on your code in src/.
Yuji from Japan,
Best,

[Yujif1Aero]

@ProjectPhysX
Calculation is done. My GPU bench mark is here.

[Yujif1Aero]

Hi @ProjectPhysX ,
I did these benchmarks,
FP16C;

FP16S;

Moreover GPU's name is here.

Best,

[ProjectPhysX]

@Yujif1Aero thank you!!

[ProjectPhysX]

Thank you for the bug report!
The OpenCL C code itself in all these places is correct, so this seems to be an issue with the C++ stringification macro being confused with brackets by your compiler. Unfortunately I can't replicate this on any of my systems. What compiler version do you use?
Try this: In src/kernel.cpp, replace )+") {"+R( // ... with )+")"+R( { // ... in lines 1278, 1385 and 1725. Let me know if it works, then I'll update it in the repository.

[BurgaoTheChamp]

I'm using this version

I'll try to update my compiler and also to do that suggestion and I'll report soon about the kernel.cpp

[BurgaoTheChamp]

Same error

btw I updated my compiler too

[PavelBlend]

I am getting the same error. I am using exe files from release:
https://github.com/ProjectPhysX/FluidX3D/releases/tag/v1.0

I don't know how to compile sources myself.

My CPU:
AMD A8-5600K APU with Radeon(tm) HD Graphics

Integrated GPU:
AMD Radeon HD 7560D ATI Technologies Inc. 4.5.13399 Core Profile Context 15.201.1151.1008

External GPU:
Missing

I run the program like this:
FluidX3D-Benchmark-FP32-FP16C-Windows.exe 0

or so:
FluidX3D-Benchmark-FP32-FP16C-Windows.exe 1

[PavelBlend]

I was able to run FluidX3D-Benchmark-FP32-FP32-Windows.exe on a GeForce GT 710.

[Shubham-wiz]

FluidX3D-Benchmark-FP32-FP32-Windows.exe working fine on amd 5600x with 3060

[ProjectPhysX]

It seems this is a driver bug with some very old AMD/ATI toasters.
With such driver bugs there sometimes is the possibility to avoid them with weird code circumventions and/or contacting the AMD driver team. But I would need to have the hardware to try&error-fix this, so unfortunately there is not really something I can do. I'll leave this thread open in case someone of you finds a solution.

[ProjectPhysX]

I've got another idea what causes this bug: the preprocessor seems to not replace the flag types with their bit patterns. Try a search-and-replace-all in src/kernel.cpp, replacing TYPE_S with 0b00000001 and so forth, for all the types listed below:

#define TYPE_S 0b00000001 // (stationary or moving) solid boundary
#define TYPE_E 0b00000010 // equilibrium boundary (inflow/outflow)
#define TYPE_T 0b00000100 // temperature boundary
#define TYPE_F 0b00001000 // fluid
#define TYPE_I 0b00010000 // interface
#define TYPE_G 0b00100000 // gas
#define TYPE_X 0b01000000 // reserved type X
#define TYPE_Y 0b10000000 // reserved type Y

#define TYPE_MS 0b00000011 // node next to moving solid boundary
#define TYPE_BO 0b00000011 // any flag bit used for boundaries (temperature excluded)
#define TYPE_IF 0b00011000 // change from interface to fluid
#define TYPE_IG 0b00110000 // change from interface to gas
#define TYPE_GI 0b00111000 // change from gas to interface
#define TYPE_SU 0b00111000 // any flag bit used for SURFACE

[ProjectPhysX]

@Yujif1Aero I have done the replacements in src/kernel.cpp. Try with this file (unzip and copy+paste): kernel.zip

[ProjectPhysX]

Hi @Yujif1Aero, thank you so much for finding the culprit! It seems that the old AMD compiler cannot handle binary-formatted number literals. Changing to hexadecimal notation fixes that. I just updated the repository!

Can you add the FP16S/FP16C benchmarks too please? Just uncomment the corresponding line in src/defines.hpp and recompile. And what exact model is your GPU? Thanks!

[Yujif1Aero]

Thank your for quick updating @ProjectPhysX.
Sure, I will add the FP16S/FP16C benchmarks and look up exact model name, after I take classes.
Wait for a while please.
I will start reading your code from tommorow. Could I ask you questions again if I struggle with reading?
Acctually, I even cannot understand your code loyout becuse your code is so amaging.

[Yujif1Aero]

@ProjectPhysX
I found exact GPU name, Radeon HD 7850