[all-commits] [llvm/llvm-project] c21724: [AArch64] Implements FP8 SVE intrinsics for dot-pr...

[Momchil Velikov via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: c2172431c72c9b249bf5bdfcc0c239fbfe64fa9b
      https://github.com/llvm/llvm-project/commit/c2172431c72c9b249bf5bdfcc0c239fbfe64fa9b
  Author: Momchil Velikov <momchil.velikov at arm.com>
  Date:   2024-12-13 (Fri, 13 Dec 2024)

  Changed paths:
    M clang/include/clang/Basic/arm_sve.td
    M clang/include/clang/Basic/arm_sve_sme_incl.td
    M clang/lib/CodeGen/CGBuiltin.cpp
    A clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sve2_fp8_fdot.c
    M clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_fp8.c
    M clang/utils/TableGen/SveEmitter.cpp
    M llvm/include/llvm/IR/IntrinsicsAArch64.td
    M llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
    M llvm/lib/Target/AArch64/SVEInstrFormats.td
    A llvm/test/CodeGen/AArch64/fp8-sve-fdot.ll

  Log Message:
  -----------
  [AArch64] Implements FP8 SVE intrinsics for dot-product (#118125)

This patch adds the following intrinsics:

* 8-bit floating-point dot product to single-precision.

// Only if (__ARM_FEATURE_SVE2 && __ARM_FEATURE_FP8DOT4) ||
__ARM_FEATURE_SSVE_FP8DOT4
svfloat32_t svdot[_f32_mf8]_fpm(svfloat32_t zda, svmfloat8_t zn,
svmfloat8_t zm, fpm_t fpm);
svfloat32_t svdot[_n_f32_mf8]_fpm(svfloat32_t zda, svmfloat8_t zn,
mfloat8_t zm, fpm_t fpm);

* 8-bit floating-point indexed dot product to single-precision.

// Only if (__ARM_FEATURE_SVE2 && __ARM_FEATURE_FP8DOT4) ||
__ARM_FEATURE_SSVE_FP8DOT4
svfloat32_t svdot_lane[_f32_mf8]_fpm(svfloat32_t zda, svmfloat8_t zn,
svmfloat8_t zm,
                                       uint64_t imm0_3, fpm_t fpm);

* 8-bit floating-point dot product to half-precision.

// Only if (__ARM_FEATURE_SVE2 && __ARM_FEATURE_FP8DOT2) ||
__ARM_FEATURE_SSVE_FP8DOT2
svfloat16_t svdot[_f16_mf8]_fpm(svfloat16_t zda, svmfloat8_t zn,
svmfloat8_t zm, fpm_t fpm);
svfloat16_t svdot[_n_f16_mf8]_fpm(svfloat16_t zda, svmfloat8_t zn,
mfloat8_t zm, fpm_t fpm);

* 8-bit floating-point indexed dot product to half-precision.

// Only if (__ARM_FEATURE_SVE2 && __ARM_FEATURE_FP8DOT2) ||
__ARM_FEATURE_SSVE_FP8DOT2
svfloat16_t svdot_lane[_f16_mf8]_fpm(svfloat16_t zda, svmfloat8_t zn,
svmfloat8_t zm,
                                       uint64_t imm0_7, fpm_t fpm);



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications