| Summary: | webgl/1.0.x/conformance/glsl/bugs/vector-matrix-constructor-scalarization.html fails | ||||||
|---|---|---|---|---|---|---|---|
| Product: | WebKit | Reporter: | Kimmo Kinnunen <kkinnunen> | ||||
| Component: | WebGL | Assignee: | Kyle Piddington <kpiddington> | ||||
| Status: | RESOLVED CONFIGURATION CHANGED | ||||||
| Severity: | Normal | CC: | dino, gman, kbr, kkinnunen, webkit-bug-importer | ||||
| Priority: | P2 | Keywords: | InRadar | ||||
| Version: | Other | ||||||
| Hardware: | Unspecified | ||||||
| OS: | Unspecified | ||||||
| Bug Depends on: | 229958 | ||||||
| Bug Blocks: | 222812, 228996 | ||||||
| Attachments: |
|
||||||
|
Description
Kimmo Kinnunen
2021-03-16 23:34:30 PDT
https://www.khronos.org/registry/webgl/sdk/tests/conformance/glsl/bugs/vector-matrix-constructor-scalarization.html?webglVersion=1&quiet=0&quick=1 Vector and matrix constructor scalarization workaround (SH_SCALARIZE_VEC_AND_MAT_CONSTRUCTOR_ARGS) caused bugs On success, you will see a series of "PASS" messages, followed by "TEST COMPLETE". Regression test for crbug.com/1165751 PASS WebGL context creation succeeded Ensure that shader translation isn't broken by the vector and matrix constructor scalarization workaround vertex shader vertex shader translated for driver fragment_shader fragment_shader translated for driver FAIL should be 5,20,0,255 at (0, 192) expected: 5,20,0,255 was 20,20,0,255 FAIL should be 45,165,0,255 at (4, 192) expected: 45,165,0,255 was 165,165,0,255 FAIL should be 83,247,0,255 at (8, 192) expected: 83,247,0,255 was 247,247,0,255 FAIL should be 120,237,0,255 at (12, 192) expected: 120,237,0,255 was 237,237,0,255 FAIL should be 153,136,0,255 at (16, 192) expected: 153,136,0,255 was 136,136,0,255 FAIL should be 183,0,0,255 at (20, 192) expected: 183,0,0,255 was 0,0,0,255 FAIL should be 208,0,0,255 at (24, 192) expected: 208,0,0,255 was 0,0,0,255 FAIL should be 229,0,0,255 at (28, 192) expected: 229,0,0,255 was 0,0,0,255 FAIL should be 243,0,0,255 at (32, 192) expected: 243,0,0,255 was 0,0,0,255 FAIL should be 252,0,0,255 at (36, 192) expected: 252,0,0,255 was 0,0,0,255 FAIL should be 255,11,0,255 at (40, 192) expected: 255,11,0,255 was 11,11,0,255 FAIL should be 251,158,0,255 at (44, 192) expected: 251,158,0,255 was 158,158,0,255 PASS should be 242,245,0,255 FAIL should be 226,240,0,255 at (52, 192) expected: 226,240,0,255 was 240,240,0,255 FAIL should be 205,143,0,255 at (56, 192) expected: 205,143,0,255 was 143,143,0,255 FAIL should be 179,0,0,255 at (60, 192) expected: 179,0,0,255 was 0,0,0,255 FAIL should be 149,0,0,255 at (64, 192) expected: 149,0,0,255 was 0,0,0,255 FAIL should be 114,0,0,255 at (68, 192) expected: 114,0,0,255 was 0,0,0,255 FAIL should be 78,0,0,255 at (72, 192) expected: 78,0,0,255 was 0,0,0,255 FAIL should be 39,0,0,255 at (76, 192) expected: 39,0,0,255 was 0,0,0,255 PASS should be 0,3,0,255 FAIL should be 0,152,0,255 at (84, 192) expected: 0,152,0,255 was 152,152,0,255 FAIL should be 0,243,0,255 at (88, 192) expected: 0,243,0,255 was 243,243,0,255 FAIL should be 0,242,0,255 at (92, 192) expected: 0,242,0,255 was 242,242,0,255 FAIL should be 0,150,0,255 at (96, 192) expected: 0,150,0,255 was 150,150,0,255 PASS should be 0,1,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 FAIL should be 0,145,0,255 at (124, 192) expected: 0,145,0,255 was 145,145,0,255 FAIL should be 0,240,0,255 at (128, 192) expected: 0,240,0,255 was 240,240,0,255 FAIL should be 0,245,0,255 at (132, 192) expected: 0,245,0,255 was 245,245,0,255 FAIL should be 0,157,0,255 at (136, 192) expected: 0,157,0,255 was 157,157,0,255 FAIL should be 0,10,0,255 at (140, 192) expected: 0,10,0,255 was 10,10,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 PASS should be 0,0,0,255 FAIL should be 36,138,0,255 at (164, 192) expected: 36,138,0,255 was 138,138,0,255 FAIL should be 75,237,0,255 at (168, 192) expected: 75,237,0,255 was 237,237,0,255 FAIL should be 112,247,0,255 at (172, 192) expected: 112,247,0,255 was 247,247,0,255 FAIL should be 146,163,0,255 at (176, 192) expected: 146,163,0,255 was 163,163,0,255 FAIL should be 177,18,0,255 at (180, 192) expected: 177,18,0,255 was 18,18,0,255 FAIL should be 203,0,0,255 at (184, 192) expected: 203,0,0,255 was 0,0,0,255 FAIL should be 225,0,0,255 at (188, 192) expected: 225,0,0,255 was 0,0,0,255 FAIL should be 241,0,0,255 at (192, 192) expected: 241,0,0,255 was 0,0,0,255 FAIL should be 251,0,0,255 at (196, 192) expected: 251,0,0,255 was 0,0,0,255 FAIL should be 255,0,0,255 at (200, 192) expected: 255,0,0,255 was 0,0,0,255 FAIL should be 253,130,0,255 at (204, 192) expected: 253,130,0,255 was 130,130,0,255 FAIL should be 244,234,0,255 at (208, 192) expected: 244,234,0,255 was 234,234,0,255 FAIL should be 230,249,0,255 at (212, 192) expected: 230,249,0,255 was 249,249,0,255 FAIL should be 210,170,0,255 at (216, 192) expected: 210,170,0,255 was 170,170,0,255 FAIL should be 185,27,0,255 at (220, 192) expected: 185,27,0,255 was 27,27,0,255 FAIL should be 155,0,0,255 at (224, 192) expected: 155,0,0,255 was 0,0,0,255 FAIL should be 122,0,0,255 at (228, 192) expected: 122,0,0,255 was 0,0,0,255 FAIL should be 86,0,0,255 at (232, 192) expected: 86,0,0,255 was 0,0,0,255 FAIL should be 47,0,0,255 at (236, 192) expected: 47,0,0,255 was 0,0,0,255 PASS should be 8,0,0,255 FAIL should be 0,123,0,255 at (244, 192) expected: 0,123,0,255 was 123,123,0,255 FAIL should be 0,230,0,255 at (248, 192) expected: 0,230,0,255 was 230,230,0,255 FAIL should be 0,251,0,255 at (252, 192) expected: 0,251,0,255 was 251,251,0,255 Running compilation test vertex shader vertex shader translated for driver fragment_shader fragment_shader translated for driver PASS Shader previously requiring SH_SCALARIZE_VEC_AND_MAT_CONSTRUCTOR_ARGS compiled successfully PASS successfullyParsed is true TEST COMPLETE *** Bug 223333 has been marked as a duplicate of this bug. *** At least macOS 11.3, iMacPro1,1, Radeon Pro Vega 56 Created attachment 435334 [details]
msl and errors
GLSL constructs something along the lines of:
mat2(vec4(1))
Which gets translated to MSL:
float2x2(float4(1))
.. which is not valid.
We need to add a cast operator and emit it?
see the attachments for the msl and errors.
GLSL:
precision mediump float;
varying vec2 vTexCoord;
float a = 0.;
#define A 0.
#define r(a) mat2( cos( a + vec4(0,-1.5708,1.5708,0) ) )
vec2 c;
#define f(U,a) ( c = (U) * r(a) , sin(10.*c.x) )
void main() {
vec2 U = vTexCoord;
gl_FragColor = U.y > .5
? vec4( f(U,a) , f(U*4.,a) , 0,1.0) // top
: vec4( f(U,A) , f(U*4.,A) , 0,1.0); // bottom
}
webkit-patch apply-from-bug 228996 git commit -a -m temporary ..then fix the issue. > We need to add a cast operator and emit it? Turns out all 24038 mat constructor operators would need a cast operator each. Maybe it would be better to fix the scalarization. Upstream bugs: https://bugs.chromium.org/p/angleproject/issues/detail?id=6306 https://bugs.chromium.org/p/angleproject/issues/detail?id=6341 https://bugs.chromium.org/p/angleproject/issues/detail?id=6340 Let's please discuss upstream on ANGLE's issue tracker. We've had fairly extensive discussions about this already with the ANGLE team on Slack (Kimmo, just added you to the thread) - I don't think we should reintroduce and fix the scalarization workaround, but instead lazily generate the needed constructors. Gregg from our team is already working on this, so let's not duplicate work. Please note that Gregg's fixed this in the direct-to-Metal translator upstream in ANGLE in: Conversions of one type to another (mat2x2->vec4) fail in direct metal backend https://bugs.chromium.org/p/angleproject/issues/detail?id=6306 and specifically: Convert constructors to function calls where needed. https://chromium-review.googlesource.com/c/angle/angle/+/3134963 Would you consider cherry-picking this fix to allow the SH_SCALARIZE_VEC_AND_MAT_CONSTRUCTOR_ARGS workaround (Bug 228996) to be disabled in WebKit's copy of ANGLE? Fixed by ANGLE update a while ago that brought Gregg's fix. |