Depal shader compile error on ATI

[unknownbrackets]

Seeing this error:

error(#272) Implicit version number 110 not supported by GL3 forward compatible context

https://report.ppsspp.org/logs/kind/970

ppsspp/GPU/Common/DepalettizeShaderCommon.cpp

Line 255 in 7f88688

WRITE(p, "#version 110\n");

I think this is a GL3 core context. We probably need to use this:

ppsspp/ext/native/gfx_es2/gpu_features.cpp

Line 79 in f8e946d

int GLExtensions::GLSLVersion() {

This appears to be new in 1.7.x, though I'm not sure why. It could just be we weren't reporting it before, but I'm marking as a regression for now.

-[Unknown]

[hrydgard]

The strange thing is that this only happens in the GenerateDepalShaderFloat path which I wouldn't expect to be used in a GL3 core context on desktop. But yes, to be safe we should definitely use the right version there anyway..

ATI btw? It's been a while since they were merged into AMD.

[unknownbrackets]

Ah, the reports all specifically had ATI as the vendor, so I assumed they were older ATI cards. But looking closer it seems to affect both AMD and ATI:

Devices affected:

• ATI Radeon HD 3200 Graphics
• ATI Radeon HD 4200
• ATI Mobility Radeon HD 4200 Series
• AMD Radeon HD 5500 Series
• AMD Radeon HD 6290 Graphics
• AMD Radeon HD 6370M
• AMD Radeon HD 6450
• AMD Radeon 6600M and 6700M Series
• AMD Radeon HD 7450
• AMD Radeon HD 7480D
• AMD Radeon HD 8470D

GL versions range from 3.3 to 4.4. It's weird that 4.4 is affected, maybe there's a bug causing it to use the float path by mistake?

-[Unknown]

[unknownbrackets]

All affected devices are sending back a slightly wrong GL_VERSION:

4.4.13294 Core Profile Context 14.502.1020.0

I wonder if we're ever getting misled by the final number there? I tried one of the version strings and we correctly parse it. Hm. Maybe somehow CheckGLExtensions being called from the wrong thread, like from CreateFramebuffer? Maybe these are people using libretro or some port with broken GL version handling?

It affects only Windows 32 and Windows 64.

-[Unknown]

[hrydgard]

Hm, this is likely fixed now, right?

[unknownbrackets]

It's not happening anymore, I'll close.

That said, I still don't really understand how it was getting into this code path on GL3. Presumably it means GPU_SUPPORTS_32BIT_INT_FSHADER is not set, but that should definitely be set on 4.x...

-[Unknown]