Finn Plummer

poly: add pass to convert poly.mul to use ntt ops

2

- We can reduce complexity of polynomial multiplication by using the number-theoretic transform - Here we add a pass that will convert all poly.mul operations into a poly.ntt of both...

Cf. https://github.com/google/heir/pull/590#r1567269081. Instead of trying to get a root for all polynomial rings at parse time, create a pass that will only materialize this parameter when needed.

Cf. Discussion following from https://github.com/google/heir/issues/543#issuecomment-2031067952. The current pre-computed (prime, 2n-th primitive root) in `scripts/generate_static_roots` pairs are just a collection of random primes for some default values of `n`. These could...

arith_ext: add transforms for arith.remui patterns

1

- We can reduce runtime computation by avoiding computing remui and instead use a Barrett reduction or SubIfGE operation when possible Resolve #710

[DXIL][SPIRV] Lower WaveActiveCountBits intrinsic

2

- add codegen for llvm builtin to spirv/directx intrinsic in CGBuiltin.cpp - add lowering of spirv intrinsic to spirv backend in SPIRVInstructionSelector.cpp - add lowering of directx intrinsic to dxil...

[HLSL][SPIRV][DXIL] Implement `dot4add_i8packed` intrinsic

1

- create a clang built-in in Builtins.td - link dot4add_i8packed in hlsl_intrinsics.h - add lowering to spirv backend through expansion of operation as OPSDot is missing up to SPIRV 1.6...

Currently we have defined tests for the 32-bit version of `countbits`, however, there is a gap for the 16-bit and 64-bit versions. https://github.com/microsoft/DirectXShaderCompiler/blob/66287b27442d0af17a152d024a6deaadb075cd30/tools/clang/unittests/HLSLExec/ShaderOpArithTable.xml#L2609 This issue tracks the addition of these...

**Description** Saw this error: ``` fatal error: generated SPIR-V is invalid: [VUID-StandaloneSpirv-Base-04781] Expected 32-bit int type for Base operand: BitCount %43 = OpBitCount %v4uint %42 ``` when compiling this code:...