Intel GPUs ,float64 type is not supported on this device

Question

Intel GPUs ,float64 type is not supported on this device

sunxiaotianmg opened this issue a year ago · comments

sunxiaotianmg commented a year ago

sunxiaotianmg · Answer 1 · Mon Sep 11 2023 13:41:00 GMT+0800 (China Standard Time)

How to handle this situation

sunxiaotianmg · Answer 2 · Mon Sep 11 2023 14:58:27 GMT+0800 (China Standard Time)

float instead of all double type ?

sunxiaotianmg · Answer 3 · Mon Sep 11 2023 16:53:23 GMT+0800 (China Standard Time)

ILGPU will query the OpenCL driver about supported functionality. In this case, OpenCL has reported that the Intel GPU does not support 64-bit floating point operations.
You will either need to stop using double, or you could potentially use the 32-bit only mode, and ILGPU will convert double to float as necesary (https://ilgpu.net/docs/03-advanced/04-math-functions/)

sunxiaotianmg · Answer 4 · Wed Sep 13 2023 16:45:25 GMT+0800 (China Standard Time)

It seems that the problem can only be solved by using float and Math (MathMode. Fast32BitOnly) together

sunxiaotianmg · Answer 5 · Wed Sep 13 2023 16:52:30 GMT+0800 (China Standard Time)

float instead of all double type，In this MathMode. Fast32BitOnly, the correct results can be obtained