Skip to content

Commit

Permalink
Merge pull request #16710 from unknownbrackets/shader-errors
Browse files Browse the repository at this point in the history 
GLES: Use uint for uint shift amounts
  • Loading branch information
@hrydgard
hrydgard authored Jan 2, 2023
2 parents fd7cc76 + 0b6635c commit 233c37a
Show file tree
Hide file tree
Showing 4 changed files with 42 additions and 36 deletions.
56 changes: 28 additions & 28 deletions GPU/Common/FragmentShaderGenerator.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -451,8 +451,8 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
}
if (enableColorTest && !colorTestAgainstZero) {
if (compat.bitwiseOps) {
WRITE(p, "uint roundAndScaleTo8x4(in vec3 x) { uvec3 u = uvec3(floor(x * 255.99)); return u.r | (u.g << 8) | (u.b << 16); }\n");
WRITE(p, "uint packFloatsTo8x4(in vec3 x) { uvec3 u = uvec3(x); return u.r | (u.g << 8) | (u.b << 16); }\n");
WRITE(p, "uint roundAndScaleTo8x4(in vec3 x) { uvec3 u = uvec3(floor(x * 255.99)); return u.r | (u.g << 0x8u) | (u.b << 0x10u); }\n");
WRITE(p, "uint packFloatsTo8x4(in vec3 x) { uvec3 u = uvec3(x); return u.r | (u.g << 0x8u) | (u.b << 0x10u); }\n");
} else if (gl_extensions.gpuVendor == GPU_VENDOR_IMGTEC) {
WRITE(p, "vec3 roundTo255thv(in vec3 x) { vec3 y = x + (0.5/255.0); return y - fract(y * 255.0) * (1.0 / 255.0); }\n");
} else {
Expand Down Expand Up @@ -497,19 +497,19 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
WRITE(p, "uint packUnorm4x8%s(%svec4 v) {\n", packSuffix, compat.shaderLanguage == GLSL_VULKAN ? "highp " : "");
WRITE(p, " highp vec4 f = clamp(v, 0.0, 1.0);\n");
WRITE(p, " uvec4 u = uvec4(255.0 * f);\n");
WRITE(p, " return u.x | (u.y << 8) | (u.z << 16) | (u.w << 24);\n");
WRITE(p, " return u.x | (u.y << 0x8u) | (u.z << 0x10u) | (u.w << 0x18u);\n");
WRITE(p, "}\n");

WRITE(p, "vec4 unpackUnorm4x8%s(highp uint x) {\n", packSuffix);
WRITE(p, " highp uvec4 u = uvec4(x & 0xFFU, (x >> 8) & 0xFFU, (x >> 16) & 0xFFU, (x >> 24) & 0xFFU);\n");
WRITE(p, " highp uvec4 u = uvec4(x & 0xFFu, (x >> 0x8u) & 0xFFu, (x >> 0x10u) & 0xFFu, (x >> 0x18u) & 0xFFu);\n");
WRITE(p, " highp vec4 f = vec4(u);\n");
WRITE(p, " return f * (1.0 / 255.0);\n");
WRITE(p, "}\n");
}

if (compat.bitwiseOps && enableColorTest) {
p.C("uvec3 unpackUVec3(highp uint x) {\n");
p.C(" return uvec3(x & 0xFFU, (x >> 8) & 0xFFU, (x >> 16) & 0xFFU);\n");
p.C(" return uvec3(x & 0xFFu, (x >> 0x8u) & 0xFFu, (x >> 0x10u) & 0xFFu);\n");
p.C("}\n");
}

Expand Down Expand Up @@ -707,8 +707,8 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
// Also, since we know the CLUT is smooth, we do not need to do the bilinear filter manually, we can just
// lookup with the filtered value once.
p.F(" vec4 t = ").SampleTexture2D("tex", "uv").C(";\n");
p.C(" uint depalShift = (u_depal_mask_shift_off_fmt >> 8) & 0xFFu;\n");
p.C(" uint depalFmt = (u_depal_mask_shift_off_fmt >> 24) & 0x3u;\n");
p.C(" uint depalShift = (u_depal_mask_shift_off_fmt >> 0x8u) & 0xFFu;\n");
p.C(" uint depalFmt = (u_depal_mask_shift_off_fmt >> 0x18u) & 0x3u;\n");
p.C(" float index0 = t.r;\n");
p.C(" float factor = 31.0 / 256.0;\n");
p.C(" if (depalFmt == 0x0u) {\n"); // yes, different versions of Test Drive use different formats. Could do compile time by adding more compat flags but meh.
Expand All @@ -730,7 +730,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
}
WRITE(p, " vec2 tsize = vec2(textureSize(tex, 0).xy);\n");
WRITE(p, " vec2 fraction;\n");
WRITE(p, " bool bilinear = (u_depal_mask_shift_off_fmt >> 31) != 0x0u;\n");
WRITE(p, " bool bilinear = (u_depal_mask_shift_off_fmt >> 0x2Fu) != 0x0u;\n");
WRITE(p, " if (bilinear) {\n");
WRITE(p, " uv_round = uv * tsize - vec2(0.5, 0.5);\n");
WRITE(p, " fraction = fract(uv_round);\n");
Expand All @@ -743,57 +743,57 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
p.C(" highp vec4 t2 = ").SampleTexture2DOffset("tex", "uv_round", 0, 1).C(";\n");
p.C(" highp vec4 t3 = ").SampleTexture2DOffset("tex", "uv_round", 1, 1).C(";\n");
WRITE(p, " uint depalMask = (u_depal_mask_shift_off_fmt & 0xFFu);\n");
WRITE(p, " uint depalShift = (u_depal_mask_shift_off_fmt >> 8) & 0xFFu;\n");
WRITE(p, " uint depalOffset = ((u_depal_mask_shift_off_fmt >> 16) & 0xFFu) << 4;\n");
WRITE(p, " uint depalFmt = (u_depal_mask_shift_off_fmt >> 24) & 0x3u;\n");
WRITE(p, " uint depalShift = (u_depal_mask_shift_off_fmt >> 0x8u) & 0xFFu;\n");
WRITE(p, " uint depalOffset = ((u_depal_mask_shift_off_fmt >> 0x10u) & 0xFFu) << 0x4u;\n");
WRITE(p, " uint depalFmt = (u_depal_mask_shift_off_fmt >> 0x18u) & 0x3u;\n");
WRITE(p, " uvec4 col; uint index0; uint index1; uint index2; uint index3;\n");
WRITE(p, " switch (int(depalFmt)) {\n"); // We might want to include fmt in the shader ID if this is a performance issue.
WRITE(p, " case 0:\n"); // 565
WRITE(p, " col = uvec4(t.rgb * vec3(31.99, 63.99, 31.99), 0);\n");
WRITE(p, " index0 = (col.b << 11) | (col.g << 5) | (col.r);\n");
WRITE(p, " index0 = (col.b << 0xBu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " if (bilinear) {\n");
WRITE(p, " col = uvec4(t1.rgb * vec3(31.99, 63.99, 31.99), 0);\n");
WRITE(p, " index1 = (col.b << 11) | (col.g << 5) | (col.r);\n");
WRITE(p, " index1 = (col.b << 0xBu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " col = uvec4(t2.rgb * vec3(31.99, 63.99, 31.99), 0);\n");
WRITE(p, " index2 = (col.b << 11) | (col.g << 5) | (col.r);\n");
WRITE(p, " index2 = (col.b << 0xBu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " col = uvec4(t3.rgb * vec3(31.99, 63.99, 31.99), 0);\n");
WRITE(p, " index3 = (col.b << 11) | (col.g << 5) | (col.r);\n");
WRITE(p, " index3 = (col.b << 0xBu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " }\n");
WRITE(p, " break;\n");
WRITE(p, " case 1:\n"); // 5551
WRITE(p, " col = uvec4(t.rgba * vec4(31.99, 31.99, 31.99, 1.0));\n");
WRITE(p, " index0 = (col.a << 15) | (col.b << 10) | (col.g << 5) | (col.r);\n");
WRITE(p, " index0 = (col.a << 0xFu) | (col.b << 0xAu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " if (bilinear) {\n");
WRITE(p, " col = uvec4(t1.rgba * vec4(31.99, 31.99, 31.99, 1.0));\n");
WRITE(p, " index1 = (col.a << 15) | (col.b << 10) | (col.g << 5) | (col.r);\n");
WRITE(p, " index1 = (col.a << 0xFu) | (col.b << 0xAu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " col = uvec4(t2.rgba * vec4(31.99, 31.99, 31.99, 1.0));\n");
WRITE(p, " index2 = (col.a << 15) | (col.b << 10) | (col.g << 5) | (col.r);\n");
WRITE(p, " index2 = (col.a << 0xFu) | (col.b << 0xAu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " col = uvec4(t3.rgba * vec4(31.99, 31.99, 31.99, 1.0));\n");
WRITE(p, " index3 = (col.a << 15) | (col.b << 10) | (col.g << 5) | (col.r);\n");
WRITE(p, " index3 = (col.a << 0xFu) | (col.b << 0xAu) | (col.g << 0x5u) | (col.r);\n");
WRITE(p, " }\n");
WRITE(p, " break;\n");
WRITE(p, " case 2:\n"); // 4444
WRITE(p, " col = uvec4(t.rgba * 15.99);\n");
WRITE(p, " index0 = (col.a << 12) | (col.b << 8) | (col.g << 4) | (col.r);\n");
WRITE(p, " index0 = (col.a << 0xCu) | (col.b << 0x8u) | (col.g << 0x4u) | (col.r);\n");
WRITE(p, " if (bilinear) {\n");
WRITE(p, " col = uvec4(t1.rgba * 15.99);\n");
WRITE(p, " index1 = (col.a << 12) | (col.b << 8) | (col.g << 4) | (col.r);\n");
WRITE(p, " index1 = (col.a << 0xCu) | (col.b << 0x8u) | (col.g << 0x4u) | (col.r);\n");
WRITE(p, " col = uvec4(t2.rgba * 15.99);\n");
WRITE(p, " index2 = (col.a << 12) | (col.b << 8) | (col.g << 4) | (col.r);\n");
WRITE(p, " index2 = (col.a << 0xCu) | (col.b << 0x8u) | (col.g << 0x4u) | (col.r);\n");
WRITE(p, " col = uvec4(t3.rgba * 15.99);\n");
WRITE(p, " index3 = (col.a << 12) | (col.b << 8) | (col.g << 4) | (col.r);\n");
WRITE(p, " index3 = (col.a << 0xCu) | (col.b << 0x8u) | (col.g << 0x4u) | (col.r);\n");
WRITE(p, " }\n");
WRITE(p, " break;\n");
WRITE(p, " case 3:\n"); // 8888
WRITE(p, " col = uvec4(t.rgba * 255.99);\n");
WRITE(p, " index0 = (col.a << 24) | (col.b << 16) | (col.g << 8) | (col.r);\n");
WRITE(p, " index0 = (col.a << 0x18u) | (col.b << 0x10u) | (col.g << 0x8u) | (col.r);\n");
WRITE(p, " if (bilinear) {\n");
WRITE(p, " col = uvec4(t1.rgba * 255.99);\n");
WRITE(p, " index1 = (col.a << 24) | (col.b << 16) | (col.g << 8) | (col.r);\n");
WRITE(p, " index1 = (col.a << 0x18u) | (col.b << 0x10u) | (col.g << 0x8u) | (col.r);\n");
WRITE(p, " col = uvec4(t2.rgba * 255.99);\n");
WRITE(p, " index2 = (col.a << 24) | (col.b << 16) | (col.g << 8) | (col.r);\n");
WRITE(p, " index2 = (col.a << 0x18u) | (col.b << 0x10u) | (col.g << 0x8u) | (col.r);\n");
WRITE(p, " col = uvec4(t3.rgba * 255.99);\n");
WRITE(p, " index3 = (col.a << 24) | (col.b << 16) | (col.g << 8) | (col.r);\n");
WRITE(p, " index3 = (col.a << 0x18u) | (col.b << 0x10u) | (col.g << 0x8u) | (col.r);\n");
WRITE(p, " }\n");
WRITE(p, " break;\n");
WRITE(p, " };\n");
Expand Down Expand Up @@ -944,7 +944,7 @@ bool GenerateFragmentShader(const FShaderID &id, char *buffer, const ShaderLangu
const char *alphaTestFuncs[] = { "#", "#", " != ", " == ", " >= ", " > ", " <= ", " < " };
if (alphaTestFuncs[alphaTestFunc][0] != '#') {
if (compat.bitwiseOps) {
WRITE(p, " if ((roundAndScaleTo255i(v.a) & int(u_alphacolormask >> 24)) %s int(u_alphacolorref >> 24)) %s\n", alphaTestFuncs[alphaTestFunc], discardStatement);
WRITE(p, " if ((roundAndScaleTo255i(v.a) & int(u_alphacolormask >> 0x18u)) %s int(u_alphacolorref >> 0x18u)) %s\n", alphaTestFuncs[alphaTestFunc], discardStatement);
} else if (gl_extensions.gpuVendor == GPU_VENDOR_IMGTEC) {
// Work around bad PVR driver problem where equality check + discard just doesn't work.
if (alphaTestFunc != GE_COMP_NOTEQUAL) {
Expand Down
5 changes: 4 additions & 1 deletion GPU/Common/TextureCacheCommon.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2131,11 +2131,14 @@ void TextureCacheCommon::ApplyTextureFramebuffer(VirtualFramebuffer *framebuffer
!gstate_c.curTextureIs3D &&
draw_->GetShaderLanguageDesc().bitwiseOps;

// TODO: Implement shader depal in the fragment shader generator for D3D11 at least.
switch (draw_->GetShaderLanguageDesc().shaderLanguage) {
case ShaderLanguage::HLSL_D3D9:
useShaderDepal = false;
break;
case ShaderLanguage::GLSL_1xx:
// Force off for now, in case <= GLSL 1.20 or GLES 2, which don't support switch-case.
useShaderDepal = false;
break;
default:
break;
}
Expand Down
10 changes: 5 additions & 5 deletions GPU/Common/VertexShaderGenerator.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -967,10 +967,10 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
}

if (lightUberShader && hasColor) {
p.F(" vec4 ambientColor = ((u_lightControl & (1u << 20u)) != 0x0u) ? %s : u_matambientalpha;\n", srcCol);
p.F(" vec4 ambientColor = ((u_lightControl & (1u << 0x14u)) != 0x0u) ? %s : u_matambientalpha;\n", srcCol);
if (enableLighting) {
p.F(" vec3 diffuseColor = ((u_lightControl & (1u << 21u)) != 0x0u) ? %s.rgb : u_matdiffuse;\n", srcCol);
p.F(" vec3 specularColor = ((u_lightControl & (1u << 22u)) != 0x0u) ? %s.rgb : u_matspecular.rgb;\n", srcCol);
p.F(" vec3 diffuseColor = ((u_lightControl & (1u << 0x15u)) != 0x0u) ? %s.rgb : u_matdiffuse;\n", srcCol);
p.F(" vec3 specularColor = ((u_lightControl & (1u << 0x16u)) != 0x0u) ? %s.rgb : u_matspecular.rgb;\n", srcCol);
}
} else {
// This path also takes care of the lightUberShader && !hasColor path, because all comparisons fail.
Expand Down Expand Up @@ -1032,8 +1032,8 @@ bool GenerateVertexShader(const VShaderID &id, char *buffer, const ShaderLanguag
// u_lightControl is computed in PackLightControlBits().
for (int i = 0; i < 4; i++) {
p.F(" if ((u_lightControl & %du) != 0x0u) { \n", 1 << i);
p.F(" uint comp = (u_lightControl >> %d) & 0x3u;\n", 4 + 4 * i);
p.F(" uint type = (u_lightControl >> %d) & 0x3u;\n", 4 + 4 * i + 2);
p.F(" uint comp = (u_lightControl >> 0x%02xu) & 0x3u;\n", 4 + 4 * i);
p.F(" uint type = (u_lightControl >> 0x%02xu) & 0x3u;\n", 4 + 4 * i + 2);
p.C(" if (type == 0x0u) {\n"); // GE_LIGHTTYPE_DIRECTIONAL
p.F(" toLight = u_lightpos%d;\n", i);
p.C(" } else {\n");
Expand Down
7 changes: 5 additions & 2 deletions GPU/GLES/GPU_GLES.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -168,11 +168,14 @@ u32 GPU_GLES::CheckGPUFeatures() const {
if (gl_extensions.ARB_texture_float || gl_extensions.OES_texture_float)
features |= GPU_USE_TEXTURE_FLOAT;

// Intel drivers have been seen rejecting fragment shader uint shifts used in the alpha test.
if (!draw_->GetShaderLanguageDesc().bitwiseOps || draw_->GetDeviceCaps().vendor == Draw::GPUVendor::VENDOR_INTEL) {
if (!draw_->GetShaderLanguageDesc().bitwiseOps) {
features |= GPU_USE_FRAGMENT_TEST_CACHE;
}

// Can't use switch-case in older glsl.
if ((gl_extensions.IsGLES && !gl_extensions.GLES3) || (!gl_extensions.IsGLES && !gl_extensions.VersionGEThan(1, 3)))
features &= ~GPU_USE_LIGHT_UBERSHADER;

if (IsVREnabled()) {
features |= GPU_USE_VIRTUAL_REALITY;
}
Expand Down

0 comments on commit 233c37a

Please sign in to comment.