Skip to content

Commit

Permalink
Fix undefined behaviour in dot product kernels
Browse files Browse the repository at this point in the history 
Signed-off-by: Clayton Smith <[email protected]>
  • Loading branch information
@argilo
argilo committed Oct 22, 2023
1 parent fe46337 commit 3bcc87d
Show file tree
Hide file tree
Showing 2 changed files with 121 additions and 168 deletions.
120 changes: 50 additions & 70 deletions kernels/volk/volk_16i_32fc_dot_prod_32fc.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -147,8 +147,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_sse(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int eighthPoints = num_points / 8;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -207,18 +206,17 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_sse(lv_32fc_t* result,
_mm_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);

number = eighthPoints * 8;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}

#endif /*LV_HAVE_SSE && LV_HAVE_MMX*/
Expand All @@ -235,8 +233,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_avx2_fma(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int sixteenthPoints = num_points / 16;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -294,22 +291,19 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_avx2_fma(lv_32fc_t* result,
_mm256_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
*realpt += dotProductVector[4];
*imagpt += dotProductVector[5];
*realpt += dotProductVector[6];
*imagpt += dotProductVector[7];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);
returnValue += lv_cmake(dotProductVector[4], dotProductVector[5]);
returnValue += lv_cmake(dotProductVector[6], dotProductVector[7]);

number = sixteenthPoints * 16;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}

#endif /*LV_HAVE_AVX2 && lV_HAVE_FMA*/
Expand All @@ -326,8 +320,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_avx2(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int sixteenthPoints = num_points / 16;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -391,22 +384,19 @@ static inline void volk_16i_32fc_dot_prod_32fc_u_avx2(lv_32fc_t* result,
_mm256_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
*realpt += dotProductVector[4];
*imagpt += dotProductVector[5];
*realpt += dotProductVector[6];
*imagpt += dotProductVector[7];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);
returnValue += lv_cmake(dotProductVector[4], dotProductVector[5]);
returnValue += lv_cmake(dotProductVector[6], dotProductVector[7]);

number = sixteenthPoints * 16;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}

#endif /*LV_HAVE_AVX2*/
Expand All @@ -424,8 +414,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_sse(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int eighthPoints = num_points / 8;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -484,18 +473,17 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_sse(lv_32fc_t* result,
_mm_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);

number = eighthPoints * 8;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}

#endif /*LV_HAVE_SSE && LV_HAVE_MMX*/
Expand All @@ -511,8 +499,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_avx2(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int sixteenthPoints = num_points / 16;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -576,22 +563,19 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_avx2(lv_32fc_t* result,
_mm256_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
*realpt += dotProductVector[4];
*imagpt += dotProductVector[5];
*realpt += dotProductVector[6];
*imagpt += dotProductVector[7];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);
returnValue += lv_cmake(dotProductVector[4], dotProductVector[5]);
returnValue += lv_cmake(dotProductVector[6], dotProductVector[7]);

number = sixteenthPoints * 16;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}


Expand All @@ -608,8 +592,7 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_avx2_fma(lv_32fc_t* result,
unsigned int number = 0;
const unsigned int sixteenthPoints = num_points / 16;

float res[2];
float *realpt = &res[0], *imagpt = &res[1];
lv_32fc_t returnValue;
const short* aPtr = input;
const float* bPtr = (float*)taps;

Expand Down Expand Up @@ -667,22 +650,19 @@ static inline void volk_16i_32fc_dot_prod_32fc_a_avx2_fma(lv_32fc_t* result,
_mm256_store_ps(dotProductVector,
dotProdVal0); // Store the results back into the dot product vector

*realpt = dotProductVector[0];
*imagpt = dotProductVector[1];
*realpt += dotProductVector[2];
*imagpt += dotProductVector[3];
*realpt += dotProductVector[4];
*imagpt += dotProductVector[5];
*realpt += dotProductVector[6];
*imagpt += dotProductVector[7];
returnValue = lv_cmake(dotProductVector[0], dotProductVector[1]);
returnValue += lv_cmake(dotProductVector[2], dotProductVector[3]);
returnValue += lv_cmake(dotProductVector[4], dotProductVector[5]);
returnValue += lv_cmake(dotProductVector[6], dotProductVector[7]);

number = sixteenthPoints * 16;
for (; number < num_points; number++) {
*realpt += ((*aPtr) * (*bPtr++));
*imagpt += ((*aPtr++) * (*bPtr++));
returnValue += lv_cmake(aPtr[0] * bPtr[0], aPtr[0] * bPtr[1]);
aPtr += 1;
bPtr += 2;
}

*result = *(lv_32fc_t*)(&res[0]);
*result = returnValue;
}


Expand Down
Loading

0 comments on commit 3bcc87d

Please sign in to comment.