1. Introduction
  2. User Guide
    1. Language Syntax
  3. API Reference
    1. General Functions
    2. Math Functions
    3. Memory Functions
    4. Misc Functions
    5. Vector Functions
    6. Types Functions
    7. All Module
    8. Csv Module
    9. Currencies Module
    10. Files Module
    11. StandardUnits Module
    12. Strings Module
    13. Web Module
    14. Intel Hardware Intrinsics
      1. Intel Aes Intrinsics
      2. Intel Avx Intrinsics
      3. Intel Avx2 Intrinsics
      4. Intel Bmi1 Intrinsics
      5. Intel Bmi1X64 Intrinsics
      6. Intel Bmi2 Intrinsics
      7. Intel Bmi2X64 Intrinsics
      8. Intel Sse Intrinsics
      9. Intel SseX64 Intrinsics
      10. Intel Sse2 Intrinsics
      11. Intel Sse2X64 Intrinsics
      12. Intel Sse3 Intrinsics
      13. Intel Ssse3 Intrinsics
      14. Intel Sse41 Intrinsics
      15. Intel Sse41X64 Intrinsics
      16. Intel Sse42 Intrinsics
      17. Intel Sse42X64 Intrinsics
    15. Arm Hardware Intrinsics
      1. Arm AdvSimd Intrinsics
      2. Arm AdvSimdArm64 Intrinsics
      3. Arm Aes Intrinsics
      4. Arm Crc32 Intrinsics
      5. Arm Crc32Arm64 Intrinsics
      6. Arm Sha1 Intrinsics
      7. Arm Sha256 Intrinsics
  4. Advanced Topics
  1. Home
  2. Documentation
  3. API Reference
  4. Intel Hardware Intrinsics

Intel Hardware Intrinsics

Module with Intel CPU Hardware intrinsics.

In order to use the functions provided by this module, you need to import this module:

>>> import HardwareIntrinsics

Usage

Let's take the example of the instruction mm_or_ps which allows to perform a OR of 4 x int32:

>>> help mm_or_ps
# mm_or_ps
#
#   Compute the bitwise OR of packed single-precision (32-bit) floating-point 
#   elements in "a" and "b", and store the results in "dst".
#
#   __m128 _mm_or_ps (__m128 a,  __m128 b)
#   ORPS xmm, xmm/m128

The arguments __m128 in the Intel documentation show that the function is expecting a 128-bit vector.

In kalk, a 128-bit vector is represented by e.g bool4, or int4 or float4 or byte16 (and more). You can pass these vectors directly to mm_or_ps:

>>> mm_or_ps(int4(1,2,3,4), int4(5,6,7,8))
# mm_or_ps(int4(1, 2, 3, 4), int4(5, 6, 7, 8))
out = float4(7E-45, 8E-45, 1E-44, 1.7E-44)

Note that the result is returned using float4. This is because m128 types is by default converted to a float4 in kalk. But you can easily bitcast it to int4:

>>> bitcast(int4, out)
# bitcast(int4, out)
out = int4(5, 6, 7, 12)

Let's use a similar example with mm_blend_ps:

>>> bitcast(int4, mm_blend_ps(int4(1,2,3,4), int4(5,6,7,8), 0b1010))
# bitcast(int4, mm_blend_ps(int4(1, 2, 3, 4), int4(5, 6, 7, 8), 10))
out = int4(1, 6, 3, 8)

Or for example using mm_extract_epi8:

>>> mm_extract_epi8(int4(1,2,3,4), 0)
# mm_extract_epi8(int4(1, 2, 3, 4), 0)
out = 1
>>> mm_extract_epi8(int4(1,2,3,4), 4)
# mm_extract_epi8(int4(1, 2, 3, 4), 4)
out = 2