Support Narrow Types in BitCast Folding Rule #4941
Conversation
This change adds support for narrow types in the BitCastScalarOrVector folding rule. According to Section 2.2.1 of the SPIR-V spec, types that are narrower than 32 bits are automatically either sign extended, or zero extended depending on the type. With that guaranteed, we should be able to use the first 32-bit word of any narrow type for the folding logic without performing any special conversions. In order to reduce code duplication, this change moves the GetU32BitValue and GetU64BitValue functions from IntConstant to ScalarConstant. Without this move, we would have needed an identical version of GetU32BitValue on FloatConstant.
|
|
| @@ -163,6 +163,21 @@ class ScalarConstant : public Constant { | |||
| return is_zero; | |||
| } | |||
|
|
|||
| uint32_t GetU32BitValue() const { | |||
| // Relies on unsigned values smaller than 32-bit being zero extended. See | |||
| // section 2.2.1 of the SPIR-V spec. | |||
There was a problem hiding this comment.
I've confirmed that the SPIR-V validator checks this rule in the LIteralsPass
(Phew!)
error: line 9: The high-order bits of a literal number in instruction 9 must be 0 for a floating-point type, or 0 for an integer type with Signedness of 0, or sign extended when Signedness is 1
%half_n0x1_bfcp_16 = OpConstant %half -0x1.bfcp+16
This change adds several new test cases to the IntegerInstructionFoldingTest which trigger the 16-bit BitCast logic. The logic for half types was also added to the integer case since we can't easily validate half float types in C++ code. It's easier to validate them as unsigned integers instead. Pllus this also allows us to verify the SPIR-V constant sign extension logic too.
|
Just pushed up some new test cases in the IntegerInstructionFoldingTest suite that cover 16-bit type casting. I handled the 16-bit floats in there too because I didn't see a clean way to integrate them into the FloatInstructionFoldingTest since it uses 32-bit C++ floats as the comparison value. I would have had to do some ugly casting to make them work there, so I figured it made more sense to treat them as integers. (Plus it also allows the 32-bit sign/zero extension logic to be tested) |
This change adds a couple more test cases to the integer instruction folding test suite in order to ensure that the BitCast logic also works correctly with the Int8 shader capability.
|
Just added two more test cases to help get some coverage for 8-bit constants. |
Hi @dneto0, just wanted to follow up on this and see if the tests I added were sufficient or if I need to add more? |
No problem! Thank you for the review! 🙂 |
This change adds support for narrow types in the BitCastScalarOrVector folding rule. According to Section 2.2.1 of the SPIR-V spec, types that are narrower than 32 bits are automatically either sign extended, or zero extended depending on the type. With that guaranteed, we should be able to use the first 32-bit word of any narrow type for the folding logic without performing any special conversions.
In order to reduce code duplication, this change moves the GetU32BitValue and GetU64BitValue functions from IntConstant to ScalarConstant. Without this move, we would have needed an identical version of GetU32BitValue on FloatConstant.