maki/openairinterface5g
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add Intrinsics usage by SIMDE

Browse Source
This commit is contained in:
Bruno Mongazon-Cazavet
2022-09-28 18:21:19 +02:00
committed by Robert Schmidt
parent a1de5e3d11
commit fe1dc24e28
114 changed files with 7900 additions and 11031 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openair1/PHY/CODING/nrSmallBlock/decodeSmallBlock.c
+13 -47
View File
@@ -53,19 +53,11 @@ uint16_t decodeSmallBlock(int8_t *in, uint8_t len){
for (int k = 0; k < NR_SMALL_BLOCK_CODED_BITS; ++k)
Rhat[j] += in[k] * hadamard32InterleavedTransposed[j][k];
#if defined(__AVX2__)
for (int i = 0; i < NR_SMALL_BLOCK_CODED_BITS; i += 16) {
__m256i a15_a0 = _mm256_loadu_si256((__m256i*)&Rhat[i]);
a15_a0 = _mm256_abs_epi16(a15_a0);
_mm256_storeu_si256((__m256i*)(&Rhatabs[i]), a15_a0);
__m256i a15_a0 = simde_mm256_loadu_si256((__m256i*)&Rhat[i]);
a15_a0 = simde_mm256_abs_epi16(a15_a0);
simde_mm256_storeu_si256((__m256i*)(&Rhatabs[i]), a15_a0);
}
#else
for (int i = 0; i < NR_SMALL_BLOCK_CODED_BITS; i += 8) {
__m128i a7_a0 = _mm_loadu_si128((__m128i*)&Rhat[i]);
a7_a0 = _mm_abs_epi16(a7_a0);
_mm_storeu_si128((__m128i*)(&Rhatabs[i]), a7_a0);
}
#endif
maxVal = Rhatabs[0];
for (int k = 1; k < jmax; ++k){
if (Rhatabs[k] > maxVal){
@@ -85,27 +77,28 @@ uint16_t decodeSmallBlock(int8_t *in, uint8_t len){
} else {
uint8_t maxRow = 0, maxCol = 0;
#if defined(__AVX2__)
int16_t maxVal = 0;
int DmatrixElementVal = 0;
#if !defined(__AVX512F__)
int8_t DmatrixElement[NR_SMALL_BLOCK_CODED_BITS] = {0};
__m256i _in_256 = _mm256_loadu_si256 ((__m256i*)&in[0]);
#endif
__m256i _in_256 = simde_mm256_loadu_si256 ((__m256i*)&in[0]);
__m256i _maskD_256, _Dmatrixj_256, _maskH_256, _DmatrixElement_256;
for (int j = 0; j < ( 1<<(len-6) ); ++j) {
_maskD_256 = _mm256_loadu_si256 ((__m256i*)(&maskD[j][0]));
_Dmatrixj_256 = _mm256_sign_epi8 (_in_256, _maskD_256);
_maskD_256 = simde_mm256_loadu_si256 ((__m256i*)(&maskD[j][0]));
_Dmatrixj_256 = simde_mm256_sign_epi8 (_in_256, _maskD_256);
for (int k = 0; k < NR_SMALL_BLOCK_CODED_BITS; ++k) {
_maskH_256 = _mm256_loadu_si256 ((__m256i*)(&hadamard32InterleavedTransposed[k][0]));
_DmatrixElement_256 = _mm256_sign_epi8 (_Dmatrixj_256, _maskH_256);
_maskH_256 = simde_mm256_loadu_si256 ((__m256i*)(&hadamard32InterleavedTransposed[k][0]));
_DmatrixElement_256 = simde_mm256_sign_epi8 (_Dmatrixj_256, _maskH_256);
#if defined(__AVX512F__)
DmatrixElementVal = _mm512_reduce_add_epi32 (
_mm512_add_epi32(
_mm512_cvtepi8_epi32 (_mm256_extracti128_si256 (_DmatrixElement_256, 1)),
_mm512_cvtepi8_epi32 (_mm256_castsi256_si128 (_DmatrixElement_256))
_mm512_cvtepi8_epi32 (simde_mm256_extracti128_si256 (_DmatrixElement_256, 1)),
_mm512_cvtepi8_epi32 (simde_mm256_castsi256_si128 (_DmatrixElement_256))
)
);
#else
_mm256_storeu_si256((__m256i*)(&DmatrixElement[0]), _DmatrixElement_256);
simde_mm256_storeu_si256((__m256i*)(&DmatrixElement[0]), _DmatrixElement_256);
for (int i = 0; i < NR_SMALL_BLOCK_CODED_BITS; ++i)
DmatrixElementVal += DmatrixElement[i];
#endif
@@ -118,33 +111,6 @@ uint16_t decodeSmallBlock(int8_t *in, uint8_t len){
}
}
out = properOrderedBasisExtended[maxRow] | properOrderedBasis[maxCol] | ( (maxVal > 0) ? (uint16_t)0 : (uint16_t)1 );
#else
int8_t Dmatrix[NR_SMALL_BLOCK_CODED_BITS][NR_SMALL_BLOCK_CODED_BITS] = {0};
int16_t DmatrixFHT[NR_SMALL_BLOCK_CODED_BITS][NR_SMALL_BLOCK_CODED_BITS] = {0};
uint16_t maxVal;
uint8_t rowLimit = 1<<(len-6);
for (int j = 0; j < ( rowLimit ); ++j)
for (int k = 0; k < NR_SMALL_BLOCK_CODED_BITS; ++k)
Dmatrix[j][k] = in[k] * maskD[j][k];
for (int i = 0; i < ( rowLimit ); ++i)
for (int j = 0; j < NR_SMALL_BLOCK_CODED_BITS; ++j)
for (int k = 0; k < NR_SMALL_BLOCK_CODED_BITS; ++k)
DmatrixFHT[i][j] += Dmatrix[i][k] * hadamard32InterleavedTransposed[j][k];
maxVal = abs(DmatrixFHT[0][0]);
for (int i = 0; i < ( rowLimit ); ++i)
for (int j = 0; j < NR_SMALL_BLOCK_CODED_BITS; ++j)
if (abs(DmatrixFHT[i][j]) > maxVal){
maxVal = abs(DmatrixFHT[i][j]);
maxRow = i;
maxCol = j;
}
out = properOrderedBasisExtended[maxRow] | properOrderedBasis[maxCol] | ( (DmatrixFHT[maxRow][maxCol] > 0) ? (uint16_t)0 : (uint16_t)1 );
#endif
#ifdef DEBUG_DECODESMALLBLOCK
for (int k = 0; k < NR_SMALL_BLOCK_CODED_BITS; ++k)
printf("[decodeSmallBlock]maxRow = %d maxCol = %d out[%d]=%d\n", maxRow, maxCol, k, ((uint32_t)out>>k)&1);