Open Jianqoq opened 1 month ago
windows11, vulkan | DX12, lastest wgpu
The SHADER_F64
feature is only implemented on Vulkan right now.
Looking at the spec https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html, there are a bunch of functions that only support "16-bit or 32-bit floating-point" numbers.
We should validate that those are only used with 16-bit or 32-bit floating-point numbers.
Description shader f64 can't compute sin, cos, tan correctly, f32 can. sinh, cosh, tanh can calculate correctly
Repro steps
Expected vs observed behavior for
[0, 1, 2, 3, 4]
f32 I can get[0. 0.84 0.90 0.14 -0.7]
, but for f64 I get[0.0, 2.121995791e-314, 4.243991582e-314, 6.365987373e-314, 8.487983164e-314]
Platform windows11, vulkan | DX12, lastest wgpu