use of avx512 at CN processing level

openair1/PHY/CODING/nrLDPC_decoder/nrLDPC_bnProc.h

@@ -2745,8 +2745,8 @@ static inline void nrLDPC_llr2bit(int8_t* out, int8_t* llrOut, uint16_t numLLR)
     uint32_t M  = numLLR>>5;
     uint32_t Mr = numLLR&31;
 
-    const __m256i* p_zeros = (__m256i*) zeros256_epi8;
-    const __m256i* p_ones  = (__m256i*) ones256_epi8;
+    const __m256i* p_zeros = (__m256i*) zeros512_epi8;
+    const __m256i* p_ones  = (__m256i*) ones512_epi8;
 
     for (i=0; i<M; i++)
     {

openair1/PHY/CODING/nrLDPC_decoder/nrLDPC_init_mem.h

@@ -34,15 +34,15 @@
 #include <stdlib.h>
 #include "nrLDPC_types.h"
 
-#ifndef malloc32_clear
+#ifndef malloc64_clear
 /**
-   \brief Allocates 32 byte aligned memory and initializes to zero
+   \brief Allocates 64 byte aligned memory and initializes to zero
    \param size Input size in bytes
    \return Pointer to memory
 */
-static inline void* malloc32_clear(size_t size)
+static inline void* malloc64_clear(size_t size)
 {
-    void* ptr = (void*) memalign(32, size+32);
+    void* ptr = (void*) memalign(64, size+64);
     memset(ptr, 0, size);
     return ptr;
 }
@@ -56,16 +56,16 @@ static inline void* malloc32_clear(size_t size)
 */
 static inline t_nrLDPC_procBuf* nrLDPC_init_mem(void)
 {
-    t_nrLDPC_procBuf* p_procBuf = (t_nrLDPC_procBuf*) malloc32_clear(sizeof(t_nrLDPC_procBuf));
+    t_nrLDPC_procBuf* p_procBuf = (t_nrLDPC_procBuf*) malloc64_clear(sizeof(t_nrLDPC_procBuf));
 
     if (p_procBuf)
     {
-        p_procBuf->cnProcBuf    = (int8_t*) malloc32_clear(NR_LDPC_SIZE_CN_PROC_BUF*sizeof(int8_t));
-        p_procBuf->cnProcBufRes = (int8_t*) malloc32_clear(NR_LDPC_SIZE_CN_PROC_BUF*sizeof(int8_t));
-        p_procBuf->bnProcBuf    = (int8_t*) malloc32_clear(NR_LDPC_SIZE_BN_PROC_BUF*sizeof(int8_t));
-        p_procBuf->bnProcBufRes = (int8_t*) malloc32_clear(NR_LDPC_SIZE_BN_PROC_BUF*sizeof(int8_t));
-        p_procBuf->llrRes       = (int8_t*) malloc32_clear(NR_LDPC_MAX_NUM_LLR     *sizeof(int8_t));
-        p_procBuf->llrProcBuf   = (int8_t*) malloc32_clear(NR_LDPC_MAX_NUM_LLR     *sizeof(int8_t));
+        p_procBuf->cnProcBuf    = (int8_t*) malloc64_clear(NR_LDPC_SIZE_CN_PROC_BUF*sizeof(int8_t));
+        p_procBuf->cnProcBufRes = (int8_t*) malloc64_clear(NR_LDPC_SIZE_CN_PROC_BUF*sizeof(int8_t));
+        p_procBuf->bnProcBuf    = (int8_t*) malloc64_clear(NR_LDPC_SIZE_BN_PROC_BUF*sizeof(int8_t));
+        p_procBuf->bnProcBufRes = (int8_t*) malloc64_clear(NR_LDPC_SIZE_BN_PROC_BUF*sizeof(int8_t));
+        p_procBuf->llrRes       = (int8_t*) malloc64_clear(NR_LDPC_MAX_NUM_LLR     *sizeof(int8_t));
+        p_procBuf->llrProcBuf   = (int8_t*) malloc64_clear(NR_LDPC_MAX_NUM_LLR     *sizeof(int8_t));
     }
 
     return(p_procBuf);

openair1/PHY/CODING/nrLDPC_decoder/nrLDPCdecoder_defs.h

@@ -191,11 +191,11 @@ static const uint16_t lut_startAddrBnGroupsLlr_BG2_R13[NR_LDPC_NUM_BN_GROUPS_BG2
 /** Start address for every BN group within the LLR processing buffer for BG2 rate = 2/3 */
 static const uint16_t lut_startAddrBnGroupsLlr_BG2_R23[NR_LDPC_NUM_BN_GROUPS_BG2_R23] = {0,  1152,  2304,  4224,  5376, 6144};
 
-/** Vector of 32 '1' in int8 for application with AVX2 */
-static const int8_t ones256_epi8[32] __attribute__ ((aligned(32))) = {1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1};
-/** Vector of 32 '0' in int8 for application with AVX2 */
-static const int8_t zeros256_epi8[32] __attribute__ ((aligned(32))) = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
-/** Vector of 32 '127' in int8 for application with AVX2 */
-static const int8_t maxLLR256_epi8[32] __attribute__ ((aligned(32))) = {127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127};
+/** Vector of 64 '1' in int8 for application with AVX512 */
+static const int8_t ones512_epi8[64] __attribute__ ((aligned(64))) = {1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1};
+/** Vector of 64 '0' in int8 for application with AVX512 */
+static const int8_t zeros512_epi8[64] __attribute__ ((aligned(64))) = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
+/** Vector of 64 '127' in int8 for application with AVX512 */
+static const int8_t maxLLR512_epi8[64] __attribute__ ((aligned(64))) = {127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127};
 
 #endif