Created
August 6, 2019 13:13
-
-
Save IJzerbaard/290e51c8df9f1e509609229a9541af67 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
void cofactorSSE(const float src[16], float dst[16]) { | |
__m128 r0 = _mm_load_ps(&src[0]); | |
__m128 r1 = _mm_load_ps(&src[4]); | |
__m128 r2 = _mm_load_ps(&src[8]); | |
__m128 r3 = _mm_load_ps(&src[12]); | |
__m128 r0_0001 = _mm_shuffle_ps(r0, r0, _MM_SHUFFLE(0, 0, 0, 1)); | |
__m128 r1_0001 = _mm_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 0, 0, 1)); | |
__m128 r2_0001 = _mm_shuffle_ps(r2, r2, _MM_SHUFFLE(0, 0, 0, 1)); | |
__m128 r3_0001 = _mm_shuffle_ps(r3, r3, _MM_SHUFFLE(0, 0, 0, 1)); | |
__m128 r0_1122 = _mm_shuffle_ps(r0, r0, _MM_SHUFFLE(1, 1, 2, 2)); | |
__m128 r1_1122 = _mm_shuffle_ps(r1, r1, _MM_SHUFFLE(1, 1, 2, 2)); | |
__m128 r2_1122 = _mm_shuffle_ps(r2, r2, _MM_SHUFFLE(1, 1, 2, 2)); | |
__m128 r3_1122 = _mm_shuffle_ps(r3, r3, _MM_SHUFFLE(1, 1, 2, 2)); | |
__m128 r0_2333 = _mm_shuffle_ps(r0, r0, _MM_SHUFFLE(2, 3, 3, 3)); | |
__m128 r1_2333 = _mm_shuffle_ps(r1, r1, _MM_SHUFFLE(2, 3, 3, 3)); | |
__m128 r2_2333 = _mm_shuffle_ps(r2, r2, _MM_SHUFFLE(2, 3, 3, 3)); | |
__m128 r3_2333 = _mm_shuffle_ps(r3, r3, _MM_SHUFFLE(2, 3, 3, 3)); | |
__m128 odd = _mm_set_ps(0.0, -0.0, 0.0, -0.0); | |
__m128 even = _mm_set_ps(-0.0, 0.0, -0.0, 0.0); | |
__m128 res0 = _mm_mul_ps(_mm_mul_ps(r1_0001, r2_1122), r3_2333); | |
__m128 res1 = _mm_mul_ps(_mm_mul_ps(r0_0001, r2_1122), r3_2333); | |
__m128 res2 = _mm_mul_ps(_mm_mul_ps(r0_0001, r1_1122), r3_2333); | |
__m128 res3 = _mm_mul_ps(_mm_mul_ps(r0_0001, r1_1122), r2_2333); | |
res0 = _mm_add_ps(res0, _mm_mul_ps(_mm_mul_ps(r1_1122, r2_2333), r3_0001)); | |
res1 = _mm_add_ps(res1, _mm_mul_ps(_mm_mul_ps(r0_1122, r2_2333), r3_0001)); | |
res2 = _mm_add_ps(res2, _mm_mul_ps(_mm_mul_ps(r0_1122, r1_2333), r3_0001)); | |
res3 = _mm_add_ps(res3, _mm_mul_ps(_mm_mul_ps(r0_1122, r1_2333), r2_0001)); | |
res0 = _mm_add_ps(res0, _mm_mul_ps(_mm_mul_ps(r1_2333, r2_0001), r3_1122)); | |
res1 = _mm_add_ps(res1, _mm_mul_ps(_mm_mul_ps(r0_2333, r2_0001), r3_1122)); | |
res2 = _mm_add_ps(res2, _mm_mul_ps(_mm_mul_ps(r0_2333, r1_0001), r3_1122)); | |
res3 = _mm_add_ps(res3, _mm_mul_ps(_mm_mul_ps(r0_2333, r1_0001), r2_1122)); | |
res0 = _mm_sub_ps(res0, _mm_mul_ps(_mm_mul_ps(r1_2333, r2_1122), r3_0001)); | |
res1 = _mm_sub_ps(res1, _mm_mul_ps(_mm_mul_ps(r0_2333, r2_1122), r3_0001)); | |
res2 = _mm_sub_ps(res2, _mm_mul_ps(_mm_mul_ps(r0_2333, r1_1122), r3_0001)); | |
res3 = _mm_sub_ps(res3, _mm_mul_ps(_mm_mul_ps(r0_2333, r1_1122), r2_0001)); | |
res0 = _mm_sub_ps(res0, _mm_mul_ps(_mm_mul_ps(r1_1122, r2_0001), r3_2333)); | |
res1 = _mm_sub_ps(res1, _mm_mul_ps(_mm_mul_ps(r0_1122, r2_0001), r3_2333)); | |
res2 = _mm_sub_ps(res2, _mm_mul_ps(_mm_mul_ps(r0_1122, r1_0001), r3_2333)); | |
res3 = _mm_sub_ps(res3, _mm_mul_ps(_mm_mul_ps(r0_1122, r1_0001), r2_2333)); | |
res0 = _mm_sub_ps(res0, _mm_mul_ps(_mm_mul_ps(r1_0001, r2_2333), r3_1122)); | |
res1 = _mm_sub_ps(res1, _mm_mul_ps(_mm_mul_ps(r0_0001, r2_2333), r3_1122)); | |
res2 = _mm_sub_ps(res2, _mm_mul_ps(_mm_mul_ps(r0_0001, r1_2333), r3_1122)); | |
res3 = _mm_sub_ps(res3, _mm_mul_ps(_mm_mul_ps(r0_0001, r1_2333), r2_1122)); | |
_mm_store_ps(&dst[0], _mm_xor_ps(res0, even)); | |
_mm_store_ps(&dst[4], _mm_xor_ps(res1, odd)); | |
_mm_store_ps(&dst[8], _mm_xor_ps(res2, even)); | |
_mm_store_ps(&dst[12], _mm_xor_ps(res3, odd)); | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment