WIP: Implement mipmap slope/const using a shader constant. #10376
Conversation
… mipmap lod bias.
…modes. Fixes the mipmap test for D3D11, Vulkan and modern GL properly, while keeping Tony Hawk games working. Also implements increasing the lod level with the render resolution properly.
| @@ -477,6 +477,8 @@ enum { | |||
| GPU_SUPPORTS_VERTEX_TEXTURE_FETCH = FLAG_BIT(11), | |||
| GPU_SUPPORTS_TEXTURE_FLOAT = FLAG_BIT(12), | |||
| GPU_SUPPORTS_16BIT_FORMATS = FLAG_BIT(13), | |||
| GPU_SUPPORTS_EXPLICIT_LOD = FLAG_BIT(14), | |||
| // 2 free bits! | |||
There was a problem hiding this comment.
Only one now right?
-[Unknown]
| @@ -1315,14 +1315,20 @@ void GPUCommon::Execute_End(u32 op, u32 diff) { | |||
| } | |||
|
|
|||
| void GPUCommon::Execute_TexLevel(u32 op, u32 diff) { | |||
| if (diff == 0xFFFFFFFF) return; | |||
|
|
|||
| if (diff == 0) | |||
There was a problem hiding this comment.
Isn't that already what FLAG_EXECUTEONCHANGE does?
-[Unknown]
| @@ -586,6 +587,31 @@ void LinkedShader::UpdateUniforms(u32 vertType, const VShaderID &vsid) { | |||
| if (dirty & DIRTY_STENCILREPLACEVALUE) { | |||
| glUniform1f(u_stencilReplaceValue, (float)gstate.getStencilTestRef() * (1.0f / 255.0f)); | |||
| } | |||
|
|
|||
| if ((dirty & DIRTY_TEXLOD) && gstate_c.Supports(GPU_SUPPORTS_EXPLICIT_LOD) && u_texlod) { | |||
There was a problem hiding this comment.
Shouldn't uniformMask handle the u_texlod (and ideally GPU_SUPPORTS_EXPLICIT_LOD)? Also, isn't -1 the invalid value?
-[Unknown]
| // We can't support more than 8 bits of frac, so truncate. | ||
| int useful = (f.u >> 15) & 0xFFFF; | ||
| // Now offset so the exponent aligns with log2f (exp=127 is 0.) | ||
| return (float)(useful - 127 * 256) * (1.0f / 256.0f); |
There was a problem hiding this comment.
Can't we just define this in terms of TexLog2 to avoid repeating code?
-[Unknown]
| @@ -61,7 +61,7 @@ | |||
| #include "Windows/GPU/WindowsVulkanContext.h" | |||
|
|
|||
| #ifdef _DEBUG | |||
| static const bool g_validate_ = true; | |||
| static const bool g_validate_ = false; | |||
There was a problem hiding this comment.
Debugging.
-[Unknown]
| @@ -27,7 +27,7 @@ | |||
| PPSSPP_EXE = None | |||
| TEST_ROOT = "pspautotests/tests/" | |||
| teamcity_mode = False | |||
| TIMEOUT = 5 | |||
| TIMEOUT = 500 | |||
There was a problem hiding this comment.
Debugging, I assume?
-[Unknown]
| float lod = 0.0; | ||
| switch (gstate.getTexLevelMode()) { | ||
| case GE_TEXLEVEL_MODE_CONST: { | ||
| float scaleLog = TexLog2F((float)gstate_c.curRTScale); |
There was a problem hiding this comment.
So this assumes that all mip levels contain the same image, just scaled, right? I guess we haven't heard it manually detected by anyone, so maybe it's true. Unless some game accidentally specifies black or something dumb for some level that never happened so was never noticed.
-[Unknown]
| @@ -86,6 +86,7 @@ enum { | |||
| FS_BIT_BLENDFUNC_B = 42, // 4 bits | |||
| FS_BIT_FLATSHADE = 46, | |||
| FS_BIT_BGRA_TEXTURE = 47, | |||
| FS_BIT_TEXLOD = 48, | |||
| // 48+ are free. | |||
|
This PR and #14789 is not related? |
|
This PR kinda failed by not solving the problems it was supposed to. Auto Max Quality just sidesteps the problem entirely instead by doing what makes sense on modern hardware and achieves better quality, so will probably not bother with working more on this anytime soon. |
Fixes the mipmap test completely in GLES3, D3D11, Vulkan without having to resort to bias hacks and range clamping.
However, we're obviously missing something because Tony Hawk is blurry, and some other games seem to choose too high a mipmap level (GTA). Really not sure it's a good idea to merge this, or what to do with it really... Might take part of it (resolution-dependent mip bias) and merge separately...
so for now, DO NOT MERGE