There appears to be a failure in conversion when going from a denormal FP32 value to FP16. For example, enter:
01B1'444A
in Hexadecimal when set to FP32. Then click on FP16, which should use the lowest value for FP16, but instead produces 0x018A, which is 0.000011742115020751953125. Small, but not low (which should be 0.0000000298023223876953125).
Interesting that when I click FP16, "Pretty" doesn't update from the previously correct value.
There appears to be a failure in conversion when going from a denormal FP32 value to FP16. For example, enter:
01B1'444A
in Hexadecimal when set to FP32. Then click on FP16, which should use the lowest value for FP16, but instead produces 0x018A, which is 0.000011742115020751953125. Small, but not low (which should be 0.0000000298023223876953125).
Interesting that when I click FP16, "Pretty" doesn't update from the previously correct value.