OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

make skylakex sgemm code more friendly for readers 

BTW some kernels were adjusted to improve performance
tags/v0.3.8^2
wjc404 GitHub 5 years ago
parent
1c67567008
commit
feaafbedd3
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 576 additions and 1071 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +467
    -0
      kernel/x86_64/sgemm_direct_skylakex.c
  2. +1
    -464
      kernel/x86_64/sgemm_kernel_16x4_skylakex.c
  3. +108
    -607
      kernel/x86_64/sgemm_kernel_16x4_skylakex_2.c

+ 467
- 0
kernel/x86_64/sgemm_direct_skylakex.c View File

@@ -0,0 +1,467 @@

/* the direct sgemm code written by Arjan van der Ven */
#include <immintrin.h>

/*
* "Direct sgemm" code. This code operates directly on the inputs and outputs
* of the sgemm call, avoiding the copies, memory realignments and threading,
* and only supports alpha = 1 and beta = 0.
* This is a common case and provides value for relatively small matrixes.
* For larger matrixes the "regular" sgemm code is superior, there the cost of
* copying/shuffling the B matrix really pays off.
*/



#define DECLARE_RESULT_512(N,M) __m512 result##N##M = _mm512_setzero_ps()
#define BROADCAST_LOAD_A_512(N,M) __m512 Aval##M = _mm512_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_512(N,M) __m512 Bval##N = _mm512_loadu_ps(&B[strideB * k + j + (N*16)])
#define MATMUL_512(N,M) result##N##M = _mm512_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_512(N,M) _mm512_storeu_ps(&R[(i+M) * strideR + j+(N*16)], result##N##M)


#define DECLARE_RESULT_256(N,M) __m256 result##N##M = _mm256_setzero_ps()
#define BROADCAST_LOAD_A_256(N,M) __m256 Aval##M = _mm256_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_256(N,M) __m256 Bval##N = _mm256_loadu_ps(&B[strideB * k + j + (N*8)])
#define MATMUL_256(N,M) result##N##M = _mm256_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_256(N,M) _mm256_storeu_ps(&R[(i+M) * strideR + j+(N*8)], result##N##M)

#define DECLARE_RESULT_128(N,M) __m128 result##N##M = _mm_setzero_ps()
#define BROADCAST_LOAD_A_128(N,M) __m128 Aval##M = _mm_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_128(N,M) __m128 Bval##N = _mm_loadu_ps(&B[strideB * k + j + (N*4)])
#define MATMUL_128(N,M) result##N##M = _mm_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_128(N,M) _mm_storeu_ps(&R[(i+M) * strideR + j+(N*4)], result##N##M)

#define DECLARE_RESULT_SCALAR(N,M) float result##N##M = 0;
#define BROADCAST_LOAD_A_SCALAR(N,M) float Aval##M = A[k + strideA * (i + M)];
#define LOAD_B_SCALAR(N,M) float Bval##N = B[k * strideB + j + N];
#define MATMUL_SCALAR(N,M) result##N##M += Aval##M * Bval##N;
#define STORE_SCALAR(N,M) R[(i+M) * strideR + j + N] = result##N##M;

int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K)
{
unsigned long long mnk = M * N * K;
/* large matrixes -> not performant */
if (mnk >= 28 * 512 * 512)
return 0;

/*
* if the B matrix is not a nice multiple if 4 we get many unaligned accesses,
* and the regular sgemm copy/realignment of data pays off much quicker
*/
if ((N & 3) != 0 && (mnk >= 8 * 512 * 512))
return 0;

#ifdef SMP
/* if we can run multithreaded, the threading changes the based threshold */
if (mnk > 2 * 350 * 512 && num_cpu_avail(3)> 1)
return 0;
#endif

return 1;
}



void sgemm_kernel_direct (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A, BLASLONG strideA, float * __restrict B, BLASLONG strideB , float * __restrict R, BLASLONG strideR)
{
int i, j, k;

int m4 = M & ~3;
int m2 = M & ~1;

int n64 = N & ~63;
int n32 = N & ~31;
int n16 = N & ~15;
int n8 = N & ~7;
int n4 = N & ~3;
int n2 = N & ~1;

i = 0;

for (i = 0; i < m4; i+=4) {

for (j = 0; j < n64; j+= 64) {
k = 0;
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2); DECLARE_RESULT_512(2, 2); DECLARE_RESULT_512(3, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3); DECLARE_RESULT_512(2, 3); DECLARE_RESULT_512(3, 3);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2); MATMUL_512(2, 2); MATMUL_512(3, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3); MATMUL_512(2, 3); MATMUL_512(3, 3);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
STORE_512(0, 2); STORE_512(1, 2); STORE_512(2, 2); STORE_512(3, 2);
STORE_512(0, 3); STORE_512(1, 3); STORE_512(2, 3); STORE_512(3, 3);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
STORE_512(0, 2); STORE_512(1, 2);
STORE_512(0, 3); STORE_512(1, 3);
}

for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
STORE_512(0, 0);
STORE_512(0, 1);
STORE_512(0, 2);
STORE_512(0, 3);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);
DECLARE_RESULT_256(0, 2);
DECLARE_RESULT_256(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);
BROADCAST_LOAD_A_256(x, 2);
BROADCAST_LOAD_A_256(x, 3);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
MATMUL_256(0, 2);
MATMUL_256(0, 3);
}
STORE_256(0, 0);
STORE_256(0, 1);
STORE_256(0, 2);
STORE_256(0, 3);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);
DECLARE_RESULT_128(0, 2);
DECLARE_RESULT_128(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);
BROADCAST_LOAD_A_128(x, 2);
BROADCAST_LOAD_A_128(x, 3);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
MATMUL_128(0, 2);
MATMUL_128(0, 3);
}
STORE_128(0, 0);
STORE_128(0, 1);
STORE_128(0, 2);
STORE_128(0, 3);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);
DECLARE_RESULT_SCALAR(0, 2); DECLARE_RESULT_SCALAR(1, 2);
DECLARE_RESULT_SCALAR(0, 3); DECLARE_RESULT_SCALAR(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);
BROADCAST_LOAD_A_SCALAR(x, 2);
BROADCAST_LOAD_A_SCALAR(x, 3);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
MATMUL_SCALAR(0, 2); MATMUL_SCALAR(1, 2);
MATMUL_SCALAR(0, 3); MATMUL_SCALAR(1, 3);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
STORE_SCALAR(0, 2); STORE_SCALAR(1, 2);
STORE_SCALAR(0, 3); STORE_SCALAR(1, 3);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0)
DECLARE_RESULT_SCALAR(0, 1)
DECLARE_RESULT_SCALAR(0, 2)
DECLARE_RESULT_SCALAR(0, 3)

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);
BROADCAST_LOAD_A_SCALAR(0, 2);
BROADCAST_LOAD_A_SCALAR(0, 3);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
MATMUL_SCALAR(0, 2);
MATMUL_SCALAR(0, 3);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
STORE_SCALAR(0, 2);
STORE_SCALAR(0, 3);
}
}

for (; i < m2; i+=2) {
j = 0;

for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
}
STORE_512(0, 0);
STORE_512(0, 1);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
}
STORE_256(0, 0);
STORE_256(0, 1);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
}
STORE_128(0, 0);
STORE_128(0, 1);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);
DECLARE_RESULT_SCALAR(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
}
}

for (; i < M; i+=1) {
j = 0;
for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
}
STORE_512(0, 0); STORE_512(1, 0);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
}
STORE_512(0, 0);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
}
STORE_256(0, 0);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
}
STORE_128(0, 0);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
LOAD_B_SCALAR(0, 0); LOAD_B_SCALAR(1, 0);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
}
STORE_SCALAR(0, 0);
}
}
}

+ 1
- 464
kernel/x86_64/sgemm_kernel_16x4_skylakex.c View File

@@ -1176,467 +1176,4 @@ CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict A, flo
return 0;
}


/*
* "Direct sgemm" code. This code operates directly on the inputs and outputs
* of the sgemm call, avoiding the copies, memory realignments and threading,
* and only supports alpha = 1 and beta = 0.
* This is a common case and provides value for relatively small matrixes.
* For larger matrixes the "regular" sgemm code is superior, there the cost of
* copying/shuffling the B matrix really pays off.
*/



#define DECLARE_RESULT_512(N,M) __m512 result##N##M = _mm512_setzero_ps()
#define BROADCAST_LOAD_A_512(N,M) __m512 Aval##M = _mm512_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_512(N,M) __m512 Bval##N = _mm512_loadu_ps(&B[strideB * k + j + (N*16)])
#define MATMUL_512(N,M) result##N##M = _mm512_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_512(N,M) _mm512_storeu_ps(&R[(i+M) * strideR + j+(N*16)], result##N##M)


#define DECLARE_RESULT_256(N,M) __m256 result##N##M = _mm256_setzero_ps()
#define BROADCAST_LOAD_A_256(N,M) __m256 Aval##M = _mm256_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_256(N,M) __m256 Bval##N = _mm256_loadu_ps(&B[strideB * k + j + (N*8)])
#define MATMUL_256(N,M) result##N##M = _mm256_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_256(N,M) _mm256_storeu_ps(&R[(i+M) * strideR + j+(N*8)], result##N##M)

#define DECLARE_RESULT_128(N,M) __m128 result##N##M = _mm_setzero_ps()
#define BROADCAST_LOAD_A_128(N,M) __m128 Aval##M = _mm_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_128(N,M) __m128 Bval##N = _mm_loadu_ps(&B[strideB * k + j + (N*4)])
#define MATMUL_128(N,M) result##N##M = _mm_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_128(N,M) _mm_storeu_ps(&R[(i+M) * strideR + j+(N*4)], result##N##M)

#define DECLARE_RESULT_SCALAR(N,M) float result##N##M = 0;
#define BROADCAST_LOAD_A_SCALAR(N,M) float Aval##M = A[k + strideA * (i + M)];
#define LOAD_B_SCALAR(N,M) float Bval##N = B[k * strideB + j + N];
#define MATMUL_SCALAR(N,M) result##N##M += Aval##M * Bval##N;
#define STORE_SCALAR(N,M) R[(i+M) * strideR + j + N] = result##N##M;

int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K)
{
unsigned long long mnk = M * N * K;
/* large matrixes -> not performant */
if (mnk >= 28 * 512 * 512)
return 0;

/*
* if the B matrix is not a nice multiple if 4 we get many unaligned accesses,
* and the regular sgemm copy/realignment of data pays off much quicker
*/
if ((N & 3) != 0 && (mnk >= 8 * 512 * 512))
return 0;

#ifdef SMP
/* if we can run multithreaded, the threading changes the based threshold */
if (mnk > 2 * 350 * 512 && num_cpu_avail(3)> 1)
return 0;
#endif

return 1;
}



void sgemm_kernel_direct (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A, BLASLONG strideA, float * __restrict B, BLASLONG strideB , float * __restrict R, BLASLONG strideR)
{
int i, j, k;

int m4 = M & ~3;
int m2 = M & ~1;

int n64 = N & ~63;
int n32 = N & ~31;
int n16 = N & ~15;
int n8 = N & ~7;
int n4 = N & ~3;
int n2 = N & ~1;

i = 0;

for (i = 0; i < m4; i+=4) {

for (j = 0; j < n64; j+= 64) {
k = 0;
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2); DECLARE_RESULT_512(2, 2); DECLARE_RESULT_512(3, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3); DECLARE_RESULT_512(2, 3); DECLARE_RESULT_512(3, 3);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2); MATMUL_512(2, 2); MATMUL_512(3, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3); MATMUL_512(2, 3); MATMUL_512(3, 3);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
STORE_512(0, 2); STORE_512(1, 2); STORE_512(2, 2); STORE_512(3, 2);
STORE_512(0, 3); STORE_512(1, 3); STORE_512(2, 3); STORE_512(3, 3);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
STORE_512(0, 2); STORE_512(1, 2);
STORE_512(0, 3); STORE_512(1, 3);
}

for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
STORE_512(0, 0);
STORE_512(0, 1);
STORE_512(0, 2);
STORE_512(0, 3);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);
DECLARE_RESULT_256(0, 2);
DECLARE_RESULT_256(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);
BROADCAST_LOAD_A_256(x, 2);
BROADCAST_LOAD_A_256(x, 3);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
MATMUL_256(0, 2);
MATMUL_256(0, 3);
}
STORE_256(0, 0);
STORE_256(0, 1);
STORE_256(0, 2);
STORE_256(0, 3);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);
DECLARE_RESULT_128(0, 2);
DECLARE_RESULT_128(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);
BROADCAST_LOAD_A_128(x, 2);
BROADCAST_LOAD_A_128(x, 3);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
MATMUL_128(0, 2);
MATMUL_128(0, 3);
}
STORE_128(0, 0);
STORE_128(0, 1);
STORE_128(0, 2);
STORE_128(0, 3);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);
DECLARE_RESULT_SCALAR(0, 2); DECLARE_RESULT_SCALAR(1, 2);
DECLARE_RESULT_SCALAR(0, 3); DECLARE_RESULT_SCALAR(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);
BROADCAST_LOAD_A_SCALAR(x, 2);
BROADCAST_LOAD_A_SCALAR(x, 3);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
MATMUL_SCALAR(0, 2); MATMUL_SCALAR(1, 2);
MATMUL_SCALAR(0, 3); MATMUL_SCALAR(1, 3);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
STORE_SCALAR(0, 2); STORE_SCALAR(1, 2);
STORE_SCALAR(0, 3); STORE_SCALAR(1, 3);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0)
DECLARE_RESULT_SCALAR(0, 1)
DECLARE_RESULT_SCALAR(0, 2)
DECLARE_RESULT_SCALAR(0, 3)

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);
BROADCAST_LOAD_A_SCALAR(0, 2);
BROADCAST_LOAD_A_SCALAR(0, 3);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
MATMUL_SCALAR(0, 2);
MATMUL_SCALAR(0, 3);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
STORE_SCALAR(0, 2);
STORE_SCALAR(0, 3);
}
}

for (; i < m2; i+=2) {
j = 0;

for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
}
STORE_512(0, 0);
STORE_512(0, 1);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
}
STORE_256(0, 0);
STORE_256(0, 1);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
}
STORE_128(0, 0);
STORE_128(0, 1);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);
DECLARE_RESULT_SCALAR(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
}
}

for (; i < M; i+=1) {
j = 0;
for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
}
STORE_512(0, 0); STORE_512(1, 0);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
}
STORE_512(0, 0);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
}
STORE_256(0, 0);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
}
STORE_128(0, 0);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
LOAD_B_SCALAR(0, 0); LOAD_B_SCALAR(1, 0);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
}
STORE_SCALAR(0, 0);
}
}
}
#include "sgemm_direct_skylakex.c"

+ 108
- 607
kernel/x86_64/sgemm_kernel_16x4_skylakex_2.c View File

@@ -1,5 +1,5 @@
/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 for k_count, %5 for c_store */
/* r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */
/* r10 to assist prefetch, r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */

#include "common.h"
#include <stdint.h>
@@ -53,26 +53,25 @@
#define SAVE_m16(ndim) SAVE_h_m16n##ndim "addq $64,%2;"
#define COMPUTE_m16(ndim) \
INIT_m16n##ndim\
"movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5;"\
"cmpq $18,%4; jb "#ndim"016162f;"\
"movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15; movq %2,%5; xorq %%r10,%%r10;"\
"cmpq $16,%4; jb "#ndim"016162f;"\
#ndim"016161:\n\t"\
"cmpq $126,%%r10; movq $126,%%r10; cmoveq %3,%%r10;"\
KERNEL_k1m16n##ndim\
KERNEL_k1m16n##ndim\
"prefetcht1 (%5); subq $63,%5; addq %%r10,%5;"\
KERNEL_k1m16n##ndim\
"prefetcht1 (%5); prefetcht1 63(%5); addq %3,%5;"\
KERNEL_k1m16n##ndim\
KERNEL_k1m16n##ndim\
KERNEL_k1m16n##ndim\
"prefetcht1 (%8); addq $32,%8;"\
"subq $6,%4; cmpq $18,%4; jnb "#ndim"016161b;"\
"prefetcht1 (%6); addq $32,%6;"\
"subq $4,%4; cmpq $16,%4; jnb "#ndim"016161b;"\
"movq %2,%5;"\
#ndim"016162:\n\t"\
"testq %4,%4; jz "#ndim"016163f;"\
"testq %4,%4; jz "#ndim"016164f;"\
#ndim"016163:\n\t"\
"prefetcht0 (%5); prefetcht0 63(%5); prefetcht0 (%5,%3,1); prefetcht0 63(%5,%3,1);"\
KERNEL_k1m16n##ndim\
"leaq (%5,%3,2),%5;"\
"decq %4; jmp "#ndim"016162b;"\
#ndim"016163:\n\t"\
"leaq (%5,%3,2),%5; decq %4; jnz "#ndim"016163b;"\
#ndim"016164:\n\t"\
"prefetcht0 (%%r14); prefetcht0 64(%%r14);"\
SAVE_m16(ndim)

@@ -212,185 +211,152 @@
#define COMPUTE_m4_n24 COMPUTE_L_m4(12,55555) COMPUTE_R_m4(12,55955)
#define COMPUTE_m4(ndim) COMPUTE_m4_n##ndim

/* m = 2 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm9 for accumulators */
/* m = 2 *//* xmm0 for alpha, xmm1-xmm3 for temporary use, xmm4-xmm15 for accumulators */
#define INIT_m2n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
#define KERNEL_k1m2n1(b_addr) \
#define KERNEL_k1m2n1 \
"vmovsd (%0),%%xmm1; addq $8,%0;"\
"vbroadcastss ("#b_addr"),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"addq $4,"#b_addr";"
#define SAVE_L_m2n1 "vmovsd (%2),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm4; vmovsd %%xmm4,(%2);"
"vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"addq $4,%1;"
#define SAVE_h_m2n1 "vmovsd (%2),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm4; vmovsd %%xmm4,(%2);"
#define INIT_m2n2 INIT_m2n1 "vpxor %%xmm5,%%xmm5,%%xmm5;"
#define KERNEL_k1m2n2(b_addr) \
#define KERNEL_k1m2n2 \
"vmovsd (%0),%%xmm1; addq $8,%0;"\
"vbroadcastss ("#b_addr"),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"vbroadcastss 4("#b_addr"),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm5;"\
"addq $8,"#b_addr";"
#define SAVE_L_m2n2 SAVE_L_m2n1 "vmovsd (%2,%3,1),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm5; vmovsd %%xmm5,(%2,%3,1);"
"vbroadcastss (%1),%%xmm2; vfmadd231ps %%xmm1,%%xmm2,%%xmm4;"\
"vbroadcastss 4(%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm5;"\
"addq $8,%1;"
#define SAVE_h_m2n2 SAVE_h_m2n1 "vmovsd (%2,%3,1),%%xmm1; vfmadd213ps %%xmm1,%%xmm0,%%xmm5; vmovsd %%xmm5,(%2,%3,1);"
#define INIT_m2n4 INIT_m2n2
#define INIT_m2n8 INIT_m2n4 "vpxor %%xmm6,%%xmm6,%%xmm6; vpxor %%xmm7,%%xmm7,%%xmm7;"
#define INIT_m2n12 INIT_m2n8 "vpxor %%xmm8,%%xmm8,%%xmm8; vpxor %%xmm9,%%xmm9,%%xmm9;"
#define KERNEL_k1m2n4(b_addr) \
"vmovups ("#b_addr"),%%xmm3; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"vbroadcastss 4(%0),%%xmm2; vfmadd231ps %%xmm3,%%xmm2,%%xmm5;"\
"addq $8,%0;"
#define KERNEL_k1m2n8(b_addr) \
"vmovups ("#b_addr"),%%xmm3; vmovups ("#b_addr",%%r12,1),%%xmm2; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4; vfmadd231ps %%xmm2,%%xmm1,%%xmm6;"\
"vbroadcastss 4(%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm5; vfmadd231ps %%xmm2,%%xmm1,%%xmm7;"\
"addq $8,%0;"
#define KERNEL_k1m2n12(b_addr) \
"vmovups ("#b_addr"),%%xmm3; vmovups ("#b_addr",%%r12,1),%%xmm2; vmovups ("#b_addr",%%r12,2),%%xmm1; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm4; vfmadd231ps %%xmm2,%%xmm10,%%xmm6; vfmadd231ps %%xmm1,%%xmm10,%%xmm8;"\
"vbroadcastss 4(%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm5; vfmadd231ps %%xmm2,%%xmm10,%%xmm7; vfmadd231ps %%xmm1,%%xmm10,%%xmm9;"\
"addq $8,%0;"
#define INIT_m2n16 INIT_m2n12 "vpxor %%xmm10,%%xmm10,%%xmm10; vpxor %%xmm11,%%xmm11,%%xmm11;"
#define INIT_m2n20 INIT_m2n16 "vpxor %%xmm12,%%xmm12,%%xmm12; vpxor %%xmm13,%%xmm13,%%xmm13;"
#define INIT_m2n24 INIT_m2n20 "vpxor %%xmm14,%%xmm14,%%xmm14; vpxor %%xmm15,%%xmm15,%%xmm15;"
#define KERNEL_h_k1m2n4 \
"vbroadcastss (%0),%%xmm1; vbroadcastss 4(%0),%%xmm2; addq $8,%0;"\
"vmovups (%1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm4; vfmadd231ps %%xmm2,%%xmm3,%%xmm5;"
#define KERNEL_k1m2n4 KERNEL_h_k1m2n4 "addq $16,%1;"
#define KERNEL_h_k1m2n8 KERNEL_h_k1m2n4 "vmovups (%1,%%r12,1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm6; vfmadd231ps %%xmm2,%%xmm3,%%xmm7;"
#define KERNEL_k1m2n8 KERNEL_h_k1m2n8 "addq $16,%1;"
#define KERNEL_k1m2n12 KERNEL_h_k1m2n8 \
"vmovups (%1,%%r12,2),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm8; vfmadd231ps %%xmm2,%%xmm3,%%xmm9; addq $16,%1;"
#define KERNEL_h_k1m2n16 KERNEL_k1m2n12 "vmovups (%%r15),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm10; vfmadd231ps %%xmm2,%%xmm3,%%xmm11;"
#define KERNEL_k1m2n16 KERNEL_h_k1m2n16 "addq $16,%%r15;"
#define KERNEL_h_k1m2n20 KERNEL_h_k1m2n16 "vmovups (%%r15,%%r12,1),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm12; vfmadd231ps %%xmm2,%%xmm3,%%xmm13;"
#define KERNEL_k1m2n20 KERNEL_h_k1m2n20 "addq $16,%%r15;"
#define KERNEL_h_k1m2n24 KERNEL_h_k1m2n20 "vmovups (%%r15,%%r12,2),%%xmm3; vfmadd231ps %%xmm1,%%xmm3,%%xmm14; vfmadd231ps %%xmm2,%%xmm3,%%xmm15;"
#define KERNEL_k1m2n24 KERNEL_h_k1m2n24 "addq $16,%%r15;"
#define unit_save_m2n4(c1,c2) \
"vunpcklps "#c2","#c1",%%xmm1; vunpckhps "#c2","#c1",%%xmm2;"\
"vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1; vmovsd %%xmm1,(%5); vmovhpd %%xmm1,(%5,%3,1);"\
"leaq (%5,%3,2),%5;"\
"vmovsd (%5),%%xmm3; vmovhpd (%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2; vmovsd %%xmm2,(%5); vmovhpd %%xmm2,(%5,%3,1);"\
"leaq (%5,%3,2),%5;"
#define SAVE_L_m2n4 "movq %2,%5;" unit_save_m2n4(%%xmm4,%%xmm5)
#define SAVE_L_m2n8 SAVE_L_m2n4 unit_save_m2n4(%%xmm6,%%xmm7)
#define SAVE_L_m2n12 SAVE_L_m2n8 unit_save_m2n4(%%xmm8,%%xmm9)
#define SAVE_R_m2n4 unit_save_m2n4(%%xmm4,%%xmm5)
#define SAVE_R_m2n8 SAVE_R_m2n4 unit_save_m2n4(%%xmm6,%%xmm7)
#define SAVE_R_m2n12 SAVE_R_m2n8 unit_save_m2n4(%%xmm8,%%xmm9)
#define COMPUTE_L_m2(ndim,sim) \
INIT_m2n##ndim\
"movq %%r13,%4; movq %%r14,%1;"\
#ndim""#sim"222:\n\t"\
"testq %4,%4; jz "#ndim""#sim"223f;"\
KERNEL_k1m2n##ndim(%1)\
"decq %4; jmp "#ndim""#sim"222b;"\
#ndim""#sim"223:\n\t"\
SAVE_L_m2n##ndim "addq $8,%2;"
#define COMPUTE_R_m2(ndim,sim) \
"salq $3,%%r13;subq %%r13,%0;sarq $3,%%r13;"\
#define SAVE_h_m2n4 "movq %2,%5;" unit_save_m2n4(%%xmm4,%%xmm5)
#define SAVE_h_m2n8 SAVE_h_m2n4 unit_save_m2n4(%%xmm6,%%xmm7)
#define SAVE_h_m2n12 SAVE_h_m2n8 unit_save_m2n4(%%xmm8,%%xmm9)
#define SAVE_h_m2n16 SAVE_h_m2n12 unit_save_m2n4(%%xmm10,%%xmm11)
#define SAVE_h_m2n20 SAVE_h_m2n16 unit_save_m2n4(%%xmm12,%%xmm13)
#define SAVE_h_m2n24 SAVE_h_m2n20 unit_save_m2n4(%%xmm14,%%xmm15)
#define SAVE_m2(ndim) SAVE_h_m2n##ndim "addq $8,%2;"
#define COMPUTE_m2(ndim) \
INIT_m2n##ndim\
"movq %%r13,%4; leaq (%%r14,%%r12,2),%%r15; addq %%r12,%%r15;"\
#ndim""#sim"222:\n\t"\
"testq %4,%4; jz "#ndim""#sim"223f;"\
KERNEL_k1m2n##ndim(%%r15)\
"decq %4; jmp "#ndim""#sim"222b;"\
#ndim""#sim"223:\n\t"\
SAVE_R_m2n##ndim
#define COMPUTE_m2_n1 COMPUTE_L_m2(1,77877)
#define COMPUTE_m2_n2 COMPUTE_L_m2(2,77877)
#define COMPUTE_m2_n4 COMPUTE_L_m2(4,77877)
#define COMPUTE_m2_n8 COMPUTE_L_m2(8,77877)
#define COMPUTE_m2_n12 COMPUTE_L_m2(12,77877)
#define COMPUTE_m2_n16 COMPUTE_L_m2(12,77777) COMPUTE_R_m2(4,77977)
#define COMPUTE_m2_n20 COMPUTE_L_m2(12,77677) COMPUTE_R_m2(8,77977)
#define COMPUTE_m2_n24 COMPUTE_L_m2(12,77577) COMPUTE_R_m2(12,77977)
#define COMPUTE_m2(ndim) COMPUTE_m2_n##ndim

/* m = 1 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm6 for accumulators */
"movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
"testq %4,%4; jz "#ndim"002022f;"\
#ndim"002021:\n\t"\
KERNEL_k1m2n##ndim "decq %4; jnz "#ndim"002021b;"\
#ndim"002022:\n\t"\
SAVE_m2(ndim)

/* m = 1 *//* xmm0 for alpha, xmm1-xmm3 and xmm10 for temporary use, xmm4-xmm9 for accumulators */
#define INIT_m1n1 "vpxor %%xmm4,%%xmm4,%%xmm4;"
#define KERNEL_k1m1n1(b_addr) \
"vmovss ("#b_addr"),%%xmm3; addq $4,"#b_addr";"\
#define KERNEL_k1m1n1 \
"vmovss (%1),%%xmm3; addq $4,%1;"\
"vmovss (%0),%%xmm1; vfmadd231ss %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#define SAVE_L_m1n1 "vfmadd213ss (%2),%%xmm0,%%xmm4; vmovss %%xmm4,(%2);"
#define SAVE_h_m1n1 "vfmadd213ss (%2),%%xmm0,%%xmm4; vmovss %%xmm4,(%2);"
#define INIT_m1n2 INIT_m1n1
#define KERNEL_k1m1n2(b_addr) \
"vmovsd ("#b_addr"),%%xmm3; addq $8,"#b_addr";"\
#define KERNEL_k1m1n2 \
"vmovsd (%1),%%xmm3; addq $8,%1;"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#define SAVE_L_m1n2 \
#define SAVE_h_m1n2 \
"vmovss (%2),%%xmm3; vinsertps $16,(%2,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm4;"\
"vmovss %%xmm4,(%2); vextractps $1,%%xmm4,(%2,%3,1);"
#define INIT_m1n4 INIT_m1n2
#define INIT_m1n8 INIT_m1n4 "vpxor %%xmm5,%%xmm5,%%xmm5;"
#define INIT_m1n12 INIT_m1n8 "vpxor %%xmm6,%%xmm6,%%xmm6;"
#define KERNEL_k1m1n4(b_addr) \
"vmovups ("#b_addr"),%%xmm3; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4;"\
"addq $4,%0;"
#define KERNEL_k1m1n8(b_addr) \
"vmovups ("#b_addr"),%%xmm3; vmovups ("#b_addr",%%r12,1),%%xmm2; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm1; vfmadd231ps %%xmm3,%%xmm1,%%xmm4; vfmadd231ps %%xmm2,%%xmm1,%%xmm5;"\
"addq $4,%0;"
#define KERNEL_k1m1n12(b_addr) \
"vmovups ("#b_addr"),%%xmm3; vmovups ("#b_addr",%%r12,1),%%xmm2; vmovups ("#b_addr",%%r12,2),%%xmm1; addq $16,"#b_addr";"\
"vbroadcastss (%0),%%xmm10; vfmadd231ps %%xmm3,%%xmm10,%%xmm4; vfmadd231ps %%xmm2,%%xmm10,%%xmm5; vfmadd231ps %%xmm1,%%xmm10,%%xmm6;"\
"addq $4,%0;"
#define INIT_m1n16 INIT_m1n12 "vpxor %%xmm7,%%xmm7,%%xmm7;"
#define INIT_m1n20 INIT_m1n16 "vpxor %%xmm8,%%xmm8,%%xmm8;"
#define INIT_m1n24 INIT_m1n20 "vpxor %%xmm9,%%xmm9,%%xmm9;"
#define KERNEL_h_k1m1n4 \
"vbroadcastss (%0),%%xmm1; addq $4,%0; vfmadd231ps (%1),%%xmm1,%%xmm4;"
#define KERNEL_k1m1n4 KERNEL_h_k1m1n4 "addq $16,%1;"
#define KERNEL_h_k1m1n8 KERNEL_h_k1m1n4 "vfmadd231ps (%1,%%r12,1),%%xmm1,%%xmm5;"
#define KERNEL_k1m1n8 KERNEL_h_k1m1n8 "addq $16,%1;"
#define KERNEL_k1m1n12 KERNEL_h_k1m1n8 "vfmadd231ps (%1,%%r12,2),%%xmm1,%%xmm6; addq $16,%1;"
#define KERNEL_h_k1m1n16 KERNEL_k1m1n12 "vfmadd231ps (%%r15),%%xmm1,%%xmm7;"
#define KERNEL_k1m1n16 KERNEL_h_k1m1n16 "addq $16,%%r15;"
#define KERNEL_h_k1m1n20 KERNEL_h_k1m1n16 "vfmadd231ps (%%r15,%%r12,1),%%xmm1,%%xmm8;"
#define KERNEL_k1m1n20 KERNEL_h_k1m1n20 "addq $16,%%r15;"
#define KERNEL_h_k1m1n24 KERNEL_h_k1m1n20 "vfmadd231ps (%%r15,%%r12,2),%%xmm1,%%xmm9;"
#define KERNEL_k1m1n24 KERNEL_h_k1m1n24 "addq $16,%%r15;"
#define unit_save_m1n4(c1) \
"vpxor %%xmm10,%%xmm10,%%xmm10; vmovsd "#c1",%%xmm10,%%xmm2; vmovhlps "#c1",%%xmm10,%%xmm1;"\
"vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm2;"\
"vmovss %%xmm2,(%5); vextractps $1,%%xmm2,(%5,%3,1); leaq (%5,%3,2),%5;"\
"vmovss (%5),%%xmm3; vinsertps $16,(%5,%3,1),%%xmm3,%%xmm3; vfmadd213ps %%xmm3,%%xmm0,%%xmm1;"\
"vmovss %%xmm1,(%5); vextractps $1,%%xmm1,(%5,%3,1); leaq (%5,%3,2),%5;"
#define SAVE_L_m1n4 "movq %2,%5;" unit_save_m1n4(%%xmm4)
#define SAVE_L_m1n8 SAVE_L_m1n4 unit_save_m1n4(%%xmm5)
#define SAVE_L_m1n12 SAVE_L_m1n8 unit_save_m1n4(%%xmm6)
#define SAVE_R_m1n4 unit_save_m1n4(%%xmm4)
#define SAVE_R_m1n8 SAVE_R_m1n4 unit_save_m1n4(%%xmm5)
#define SAVE_R_m1n12 SAVE_R_m1n8 unit_save_m1n4(%%xmm6)
#define COMPUTE_L_m1(ndim,sim) \
INIT_m1n##ndim\
"movq %%r13,%4; movq %%r14,%1;"\
#ndim""#sim"112:\n\t"\
"testq %4,%4; jz "#ndim""#sim"113f;"\
KERNEL_k1m1n##ndim(%1)\
"decq %4; jmp "#ndim""#sim"112b;"\
#ndim""#sim"113:\n\t"\
SAVE_L_m1n##ndim "addq $4,%2;"
#define COMPUTE_R_m1(ndim,sim) \
"salq $2,%%r13;subq %%r13,%0;sarq $2,%%r13;"\
#define SAVE_h_m1n4 "movq %2,%5;" unit_save_m1n4(%%xmm4)
#define SAVE_h_m1n8 SAVE_h_m1n4 unit_save_m1n4(%%xmm5)
#define SAVE_h_m1n12 SAVE_h_m1n8 unit_save_m1n4(%%xmm6)
#define SAVE_h_m1n16 SAVE_h_m1n12 unit_save_m1n4(%%xmm7)
#define SAVE_h_m1n20 SAVE_h_m1n16 unit_save_m1n4(%%xmm8)
#define SAVE_h_m1n24 SAVE_h_m1n20 unit_save_m1n4(%%xmm9)
#define SAVE_m1(ndim) SAVE_h_m1n##ndim "addq $4,%2;"
#define COMPUTE_m1(ndim) \
INIT_m1n##ndim\
"movq %%r13,%4; leaq (%%r14,%%r12,2),%%r15; addq %%r12,%%r15;"\
#ndim""#sim"112:\n\t"\
"testq %4,%4; jz "#ndim""#sim"113f;"\
KERNEL_k1m1n##ndim(%%r15)\
"decq %4; jmp "#ndim""#sim"112b;"\
#ndim""#sim"113:\n\t"\
SAVE_R_m1n##ndim
#define COMPUTE_m1_n1 COMPUTE_L_m1(1,99899)
#define COMPUTE_m1_n2 COMPUTE_L_m1(2,99899)
#define COMPUTE_m1_n4 COMPUTE_L_m1(4,99899)
#define COMPUTE_m1_n8 COMPUTE_L_m1(8,99899)
#define COMPUTE_m1_n12 COMPUTE_L_m1(12,99899)
#define COMPUTE_m1_n16 COMPUTE_L_m1(12,99799) COMPUTE_R_m1(4,99999)
#define COMPUTE_m1_n20 COMPUTE_L_m1(12,99699) COMPUTE_R_m1(8,99999)
#define COMPUTE_m1_n24 COMPUTE_L_m1(12,99599) COMPUTE_R_m1(12,99999)
#define COMPUTE_m1(ndim) COMPUTE_m1_n##ndim
"movq %%r13,%4; movq %%r14,%1; leaq (%1,%%r12,2),%%r15; addq %%r12,%%r15;"\
"testq %4,%4; jz "#ndim"001012f;"\
#ndim"001011:\n\t"\
KERNEL_k1m1n##ndim "decq %4; jnz "#ndim"001011b;"\
#ndim"001012:\n\t"\
SAVE_m1(ndim)

/* %0 = "+r"(a_pointer), %1 = "+r"(b_pointer), %2 = "+r"(c_pointer), %3 = "+r"(ldc_in_bytes), %4 = "+r"(K), %5 = "+r"(ctemp) */
/* %6 = "+r"(&alpha), %7 = "+r"(M), %8 = "+r"(next_b) */
/* r11 = m(const), r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */
/* %6 = "+r"(next_b), %7 = "m"(ALPHA), %8 = "m"(M) */
/* r11 = m_counter, r12 = k << 4(const), r13 = k(const), r14 = b_head_pos(const), r15 = %1 + 3r12 */

#define COMPUTE(ndim) {\
next_b = b_pointer + ndim * K;\
__asm__ __volatile__(\
"vbroadcastss (%6),%%zmm0;"\
"movq %4,%%r13; movq %4,%%r12; salq $4,%%r12; movq %1,%%r14; movq %7,%%r11;"\
"cmpq $16,%7;jb 33101"#ndim"f;"\
"vbroadcastss %7,%%zmm0;"\
"movq %4,%%r13; movq %4,%%r12; salq $4,%%r12; movq %1,%%r14; movq %8,%%r11;"\
"cmpq $16,%%r11;jb 33101"#ndim"f;"\
"33109"#ndim":\n\t"\
COMPUTE_m16(ndim)\
"subq $16,%7;cmpq $16,%7;jnb 33109"#ndim"b;"\
"subq $16,%%r11;cmpq $16,%%r11;jnb 33109"#ndim"b;"\
"33101"#ndim":\n\t"\
"cmpq $8,%7;jb 33102"#ndim"f;"\
"cmpq $8,%%r11;jb 33102"#ndim"f;"\
COMPUTE_m8(ndim)\
"subq $8,%7;"\
"subq $8,%%r11;"\
"33102"#ndim":\n\t"\
"cmpq $4,%7;jb 33103"#ndim"f;"\
"cmpq $4,%%r11;jb 33103"#ndim"f;"\
COMPUTE_m4(ndim)\
"subq $4,%7;"\
"subq $4,%%r11;"\
"33103"#ndim":\n\t"\
"cmpq $2,%7;jb 33104"#ndim"f;"\
"cmpq $2,%%r11;jb 33104"#ndim"f;"\
COMPUTE_m2(ndim)\
"subq $2,%7;"\
"subq $2,%%r11;"\
"33104"#ndim":\n\t"\
"testq %7,%7;jz 33105"#ndim"f;"\
"testq %%r11,%%r11;jz 33105"#ndim"f;"\
COMPUTE_m1(ndim)\
"33105"#ndim":\n\t"\
"movq %%r13,%4; movq %%r14,%1; movq %%r11,%7;"\
:"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(alp),"+r"(M),"+r"(next_b)\
::"r11","r12","r13","r14","r15","zmm0","zmm1","zmm2","zmm3","zmm4","zmm5","zmm6","zmm7","zmm8","zmm9","zmm10","zmm11","zmm12","zmm13","zmm14",\
"movq %%r13,%4; movq %%r14,%1; vzeroupper;"\
:"+r"(a_pointer),"+r"(b_pointer),"+r"(c_pointer),"+r"(ldc_in_bytes),"+r"(K),"+r"(ctemp),"+r"(next_b):"m"(ALPHA),"m"(M)\
:"r10","r11","r12","r13","r14","r15","zmm0","zmm1","zmm2","zmm3","zmm4","zmm5","zmm6","zmm7","zmm8","zmm9","zmm10","zmm11","zmm12","zmm13","zmm14",\
"zmm15","zmm16","zmm17","zmm18","zmm19","zmm20","zmm21","zmm22","zmm23","zmm24","zmm25","zmm26","zmm27","zmm28","zmm29","zmm30","zmm31",\
"cc","memory");\
a_pointer -= M * K; b_pointer += ndim * K;c_pointer += LDC * ndim - M;\
a_pointer -= M * K; b_pointer += ndim * K; c_pointer += LDC * ndim - M;\
}
int __attribute__ ((noinline))
CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict__ A, float * __restrict__ B, float * __restrict__ C, BLASLONG LDC)
@@ -399,7 +365,7 @@ CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict__ A, f
int64_t ldc_in_bytes = (int64_t)LDC * sizeof(float);float ALPHA = alpha;
int64_t M = (int64_t)m, K = (int64_t)k;
BLASLONG n_count = n;
float *a_pointer = A,*b_pointer = B,*c_pointer = C,*ctemp = C,*alp = &ALPHA,*next_b = B;
float *a_pointer = A,*b_pointer = B,*c_pointer = C,*ctemp = C,*next_b = B;
for(;n_count>23;n_count-=24) COMPUTE(24)
for(;n_count>19;n_count-=20) COMPUTE(20)
for(;n_count>15;n_count-=16) COMPUTE(16)
@@ -411,469 +377,4 @@ CNAME(BLASLONG m, BLASLONG n, BLASLONG k, float alpha, float * __restrict__ A, f
return 0;
}

#include <immintrin.h>
/* codes below are copied from the sgemm kernel written by Arjan van der Ven */

/*
* "Direct sgemm" code. This code operates directly on the inputs and outputs
* of the sgemm call, avoiding the copies, memory realignments and threading,
* and only supports alpha = 1 and beta = 0.
* This is a common case and provides value for relatively small matrixes.
* For larger matrixes the "regular" sgemm code is superior, there the cost of
* copying/shuffling the B matrix really pays off.
*/



#define DECLARE_RESULT_512(N,M) __m512 result##N##M = _mm512_setzero_ps()
#define BROADCAST_LOAD_A_512(N,M) __m512 Aval##M = _mm512_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_512(N,M) __m512 Bval##N = _mm512_loadu_ps(&B[strideB * k + j + (N*16)])
#define MATMUL_512(N,M) result##N##M = _mm512_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_512(N,M) _mm512_storeu_ps(&R[(i+M) * strideR + j+(N*16)], result##N##M)


#define DECLARE_RESULT_256(N,M) __m256 result##N##M = _mm256_setzero_ps()
#define BROADCAST_LOAD_A_256(N,M) __m256 Aval##M = _mm256_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_256(N,M) __m256 Bval##N = _mm256_loadu_ps(&B[strideB * k + j + (N*8)])
#define MATMUL_256(N,M) result##N##M = _mm256_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_256(N,M) _mm256_storeu_ps(&R[(i+M) * strideR + j+(N*8)], result##N##M)

#define DECLARE_RESULT_128(N,M) __m128 result##N##M = _mm_setzero_ps()
#define BROADCAST_LOAD_A_128(N,M) __m128 Aval##M = _mm_broadcastss_ps(_mm_load_ss(&A[k + strideA * (i+M)]))
#define LOAD_B_128(N,M) __m128 Bval##N = _mm_loadu_ps(&B[strideB * k + j + (N*4)])
#define MATMUL_128(N,M) result##N##M = _mm_fmadd_ps(Aval##M, Bval##N , result##N##M)
#define STORE_128(N,M) _mm_storeu_ps(&R[(i+M) * strideR + j+(N*4)], result##N##M)

#define DECLARE_RESULT_SCALAR(N,M) float result##N##M = 0;
#define BROADCAST_LOAD_A_SCALAR(N,M) float Aval##M = A[k + strideA * (i + M)];
#define LOAD_B_SCALAR(N,M) float Bval##N = B[k * strideB + j + N];
#define MATMUL_SCALAR(N,M) result##N##M += Aval##M * Bval##N;
#define STORE_SCALAR(N,M) R[(i+M) * strideR + j + N] = result##N##M;

int sgemm_kernel_direct_performant(BLASLONG M, BLASLONG N, BLASLONG K)
{
unsigned long long mnk = M * N * K;
/* large matrixes -> not performant */
if (mnk >= 28 * 512 * 512)
return 0;

/*
* if the B matrix is not a nice multiple if 4 we get many unaligned accesses,
* and the regular sgemm copy/realignment of data pays off much quicker
*/
if ((N & 3) != 0 && (mnk >= 8 * 512 * 512))
return 0;

#ifdef SMP
/* if we can run multithreaded, the threading changes the based threshold */
if (mnk > 2 * 350 * 512 && num_cpu_avail(3)> 1)
return 0;
#endif

return 1;
}



void sgemm_kernel_direct (BLASLONG M, BLASLONG N, BLASLONG K, float * __restrict A, BLASLONG strideA, float * __restrict B, BLASLONG strideB , float * __restrict R, BLASLONG strideR)
{
int i, j, k;

int m4 = M & ~3;
int m2 = M & ~1;

int n64 = N & ~63;
int n32 = N & ~31;
int n16 = N & ~15;
int n8 = N & ~7;
int n4 = N & ~3;
int n2 = N & ~1;

i = 0;

for (i = 0; i < m4; i+=4) {

for (j = 0; j < n64; j+= 64) {
k = 0;
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2); DECLARE_RESULT_512(2, 2); DECLARE_RESULT_512(3, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3); DECLARE_RESULT_512(2, 3); DECLARE_RESULT_512(3, 3);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2); MATMUL_512(2, 2); MATMUL_512(3, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3); MATMUL_512(2, 3); MATMUL_512(3, 3);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
STORE_512(0, 2); STORE_512(1, 2); STORE_512(2, 2); STORE_512(3, 2);
STORE_512(0, 3); STORE_512(1, 3); STORE_512(2, 3); STORE_512(3, 3);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
STORE_512(0, 2); STORE_512(1, 2);
STORE_512(0, 3); STORE_512(1, 3);
}

for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);
BROADCAST_LOAD_A_512(x, 2);
BROADCAST_LOAD_A_512(x, 3);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
STORE_512(0, 0);
STORE_512(0, 1);
STORE_512(0, 2);
STORE_512(0, 3);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);
DECLARE_RESULT_256(0, 2);
DECLARE_RESULT_256(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);
BROADCAST_LOAD_A_256(x, 2);
BROADCAST_LOAD_A_256(x, 3);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
MATMUL_256(0, 2);
MATMUL_256(0, 3);
}
STORE_256(0, 0);
STORE_256(0, 1);
STORE_256(0, 2);
STORE_256(0, 3);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);
DECLARE_RESULT_128(0, 2);
DECLARE_RESULT_128(0, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);
BROADCAST_LOAD_A_128(x, 2);
BROADCAST_LOAD_A_128(x, 3);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
MATMUL_128(0, 2);
MATMUL_128(0, 3);
}
STORE_128(0, 0);
STORE_128(0, 1);
STORE_128(0, 2);
STORE_128(0, 3);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);
DECLARE_RESULT_SCALAR(0, 2); DECLARE_RESULT_SCALAR(1, 2);
DECLARE_RESULT_SCALAR(0, 3); DECLARE_RESULT_SCALAR(1, 3);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);
BROADCAST_LOAD_A_SCALAR(x, 2);
BROADCAST_LOAD_A_SCALAR(x, 3);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
MATMUL_SCALAR(0, 2); MATMUL_SCALAR(1, 2);
MATMUL_SCALAR(0, 3); MATMUL_SCALAR(1, 3);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
STORE_SCALAR(0, 2); STORE_SCALAR(1, 2);
STORE_SCALAR(0, 3); STORE_SCALAR(1, 3);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0)
DECLARE_RESULT_SCALAR(0, 1)
DECLARE_RESULT_SCALAR(0, 2)
DECLARE_RESULT_SCALAR(0, 3)

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);
BROADCAST_LOAD_A_SCALAR(0, 2);
BROADCAST_LOAD_A_SCALAR(0, 3);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
MATMUL_SCALAR(0, 2);
MATMUL_SCALAR(0, 3);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
STORE_SCALAR(0, 2);
STORE_SCALAR(0, 3);
}
}

for (; i < m2; i+=2) {
j = 0;

for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(2, 1); DECLARE_RESULT_512(3, 1);


for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);

MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1); MATMUL_512(2, 1); MATMUL_512(3, 1);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
STORE_512(0, 1); STORE_512(1, 1); STORE_512(2, 1); STORE_512(3, 1);
}

for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x); LOAD_B_512(1, x);

MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
BROADCAST_LOAD_A_512(x, 1);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
MATMUL_512(0, 1);
}
STORE_512(0, 0);
STORE_512(0, 1);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);
DECLARE_RESULT_256(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
BROADCAST_LOAD_A_256(x, 1);

LOAD_B_256(0, x);

MATMUL_256(0, 0);
MATMUL_256(0, 1);
}
STORE_256(0, 0);
STORE_256(0, 1);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);
DECLARE_RESULT_128(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
BROADCAST_LOAD_A_128(x, 1);

LOAD_B_128(0, x);

MATMUL_128(0, 0);
MATMUL_128(0, 1);
}
STORE_128(0, 0);
STORE_128(0, 1);
}
for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);
DECLARE_RESULT_SCALAR(0, 1); DECLARE_RESULT_SCALAR(1, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
BROADCAST_LOAD_A_SCALAR(x, 1);

LOAD_B_SCALAR(0, x); LOAD_B_SCALAR(1, x);

MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
MATMUL_SCALAR(0, 1); MATMUL_SCALAR(1, 1);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
STORE_SCALAR(0, 1); STORE_SCALAR(1, 1);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);
DECLARE_RESULT_SCALAR(0, 1);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
BROADCAST_LOAD_A_SCALAR(0, 1);

LOAD_B_SCALAR(0, 0);

MATMUL_SCALAR(0, 0);
MATMUL_SCALAR(0, 1);
}
STORE_SCALAR(0, 0);
STORE_SCALAR(0, 1);
}
}

for (; i < M; i+=1) {
j = 0;
for (; j < n64; j+= 64) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(2, 0); DECLARE_RESULT_512(3, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x); LOAD_B_512(2, x); LOAD_B_512(3, x);
MATMUL_512(0, 0); MATMUL_512(1, 0); MATMUL_512(2, 0); MATMUL_512(3, 0);
}
STORE_512(0, 0); STORE_512(1, 0); STORE_512(2, 0); STORE_512(3, 0);
}
for (; j < n32; j+= 32) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);
LOAD_B_512(0, x); LOAD_B_512(1, x);
MATMUL_512(0, 0); MATMUL_512(1, 0);
}
STORE_512(0, 0); STORE_512(1, 0);
}


for (; j < n16; j+= 16) {
DECLARE_RESULT_512(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_512(x, 0);

LOAD_B_512(0, x);

MATMUL_512(0, 0);
}
STORE_512(0, 0);
}

for (; j < n8; j+= 8) {
DECLARE_RESULT_256(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_256(x, 0);
LOAD_B_256(0, x);
MATMUL_256(0, 0);
}
STORE_256(0, 0);
}

for (; j < n4; j+= 4) {
DECLARE_RESULT_128(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_128(x, 0);
LOAD_B_128(0, x);
MATMUL_128(0, 0);
}
STORE_128(0, 0);
}

for (; j < n2; j+= 2) {
DECLARE_RESULT_SCALAR(0, 0); DECLARE_RESULT_SCALAR(1, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(x, 0);
LOAD_B_SCALAR(0, 0); LOAD_B_SCALAR(1, 0);
MATMUL_SCALAR(0, 0); MATMUL_SCALAR(1, 0);
}
STORE_SCALAR(0, 0); STORE_SCALAR(1, 0);
}

for (; j < N; j++) {
DECLARE_RESULT_SCALAR(0, 0);

for (k = 0; k < K; k++) {
BROADCAST_LOAD_A_SCALAR(0, 0);
LOAD_B_SCALAR(0, 0);
MATMUL_SCALAR(0, 0);
}
STORE_SCALAR(0, 0);
}
}
}
#include "sgemm_direct_skylakex.c"

Write Preview
Loading…
Cancel
Save