odin-blend2d

Odin bindings to Blend2D
Log | Files | Refs | README | LICENSE

asmjit_test_unicompiler_sse2.cpp (2105B)

      1 // This file is part of AsmJit project <https://asmjit.com>
      2 //
      3 // See <asmjit/core.h> or LICENSE.md for license and copyright information
      4 // SPDX-License-Identifier: Zlib
      5 
      6 #include <asmjit/ujit.h>
      7 
      8 #if defined(ASMJIT_UJIT_X86)
      9 
     10 #if defined(_MSC_VER)
     11   #include <intrin.h>
     12 #else
     13   #include <emmintrin.h>
     14 #endif
     15 
     16 namespace UniCompilerTests {
     17 
     18 // A reference implementation of MUL+ADD without the use of FMA. This has to be provided otherwise the
     19 // compiler may use FPU registers in 32-bit x86 case, which would make the result different than when
     20 // compiled by JIT compiler that would use XMM registers (32/64-bit SSE/AVX operations).
     21 
     22 float fadd(float a, float b) noexcept {
     23   return _mm_cvtss_f32(_mm_add_ss(_mm_set1_ps(a), _mm_set1_ps(b)));
     24 }
     25 
     26 float fsub(float a, float b) noexcept {
     27   return _mm_cvtss_f32(_mm_sub_ss(_mm_set1_ps(a), _mm_set1_ps(b)));
     28 }
     29 
     30 float fmul(float a, float b) noexcept {
     31   return _mm_cvtss_f32(_mm_mul_ss(_mm_set1_ps(a), _mm_set1_ps(b)));
     32 }
     33 
     34 float fdiv(float a, float b) noexcept {
     35   return _mm_cvtss_f32(_mm_div_ss(_mm_set1_ps(a), _mm_set1_ps(b)));
     36 }
     37 
     38 float fsqrt(float a) noexcept {
     39   return _mm_cvtss_f32(_mm_sqrt_ss(_mm_set1_ps(a)));
     40 }
     41 
     42 float fmadd_nofma_ref(float a, float b, float c) noexcept {
     43   return _mm_cvtss_f32(_mm_add_ss(_mm_mul_ss(_mm_set1_ps(a), _mm_set1_ps(b)), _mm_set1_ps(c)));
     44 }
     45 
     46 double fadd(double a, double b) noexcept {
     47   return _mm_cvtsd_f64(_mm_add_sd(_mm_set1_pd(a), _mm_set1_pd(b)));
     48 }
     49 
     50 double fsub(double a, double b) noexcept {
     51   return _mm_cvtsd_f64(_mm_sub_sd(_mm_set1_pd(a), _mm_set1_pd(b)));
     52 }
     53 
     54 double fmul(double a, double b) noexcept {
     55   return _mm_cvtsd_f64(_mm_mul_sd(_mm_set1_pd(a), _mm_set1_pd(b)));
     56 }
     57 
     58 double fdiv(double a, double b) noexcept {
     59   return _mm_cvtsd_f64(_mm_div_sd(_mm_set1_pd(a), _mm_set1_pd(b)));
     60 }
     61 
     62 double fsqrt(double a) noexcept {
     63   return _mm_cvtsd_f64(_mm_sqrt_sd(_mm_setzero_pd(), _mm_set1_pd(a)));
     64 }
     65 
     66 double fmadd_nofma_ref(double a, double b, double c) noexcept {
     67   return _mm_cvtsd_f64(_mm_add_sd(_mm_mul_sd(_mm_set1_pd(a), _mm_set1_pd(b)), _mm_set1_pd(c)));
     68 }
     69 
     70 } // {UniCompilerTests}
     71 
     72 #endif // ASMJIT_UJIT_X86