DGEMM: SIMD - CS 61C Course Notes

1Learning Outcomes¶

Compare SIMD-extended matrix multiplication with sequential programs.
Compare performance using different gcc optimization flags: -O0, -O2, -O3.

🎥 Lecture Video

Before continuing, we recommend reviewing the DGEMM benchmark and the performance optimizations we tried on a SISD architecture.

DGEMM for 8 \times 8 square matrices A and B. C_{00} is computed as the dot product of row i = 0 of A and column j = 0 of B. — Figure 4:DGEMM for $8 \times 8$ square matrices A and B. $C_{00}$ is computed as the dot product of row `i = 0` of A and column `j = 0` of B.

2DGEMM 7: Naive SIMD DGEMM¶

With the ability to perform SIMD multiplication, it is tempting to use our new 256-bit-wide registers to load in blocks of row A and blocks of column B for the element-wise multiplication necessary for dot products, as shown in Figure 1.

"TODO" — Figure 1:“Naive” SIMD DGEMM that leverages SIMD architecture registers to parallelize multiplications *within* a single dot product. The outlined boxes indicate which values are loaded into the 256-bit-wide registers.^[1]

This naive SIMD implementation does better than our naive DGEMM but worse than specifying narrow registers with the register keyword in C:

C               0.768672 seconds
registers:      0.277462 seconds
simd,naive:     0.416584 seconds

The non-compulsory cache miss persists in our naive SIMD implementation—we are still loading in multiple rows of B instead of columns of B.

3DGEMM 8: Transpose SIMD DGEMM¶

We next apply cache blocking by first transposing B, then leveraging our 256-bit-wide (four-double) registers for element-wise multiplication:

C               0.768672 seconds
registers:      0.277462 seconds
simd,naive:     0.416584 seconds
simd,transpose: 0.275622 seconds

This version is certainly speedier, but it does not prove huge benefits beyond our register keyword approach.

4DGEMM 9: Tiled SIMD DGEMM¶

Recall that cache blocking is any re-design of our algorithm to adjust memory accesses. Earlier, we discussed a submatrix tiling approach to matrix multiplication—where we compute multiple elements of C with the current set of rows of A and set of columns of B.

In Figure 3, we assume that the product of a scalar with a vector can be computed as a vector operation, provided that the scalar is copied to each element in a vector of the same length.

We can extend this idea to a tiled SIMD approach shown in Figure 4.

Figure 4:SIMD dgemm tiled matrix multiplication. The outlined boxes indicate which values are loaded into the 256-bit-wide registers. Use the menu bar to trace through the animation or access the original Google Slides.

Show Explanation

Figure 4 shows an 8-by-8 matrix multiplication.

i = 0. Compute elements $C_{0j}$ for $j = 0, 1, 2, 3$ ; let this be $\vec{c}$ . Let $\vec{b}$ be the corresponding part of the $k$ -th row of $B$ .
1a through 1h: Loop through $k = 0, ..., 7$ . Compute the result of $A_{ik} \cdot$ \vec{b} $and add to the four elements of$ \vec{c}$.
i = 0, still. Next, compute elements $C_{0j}$ for $j = 4, 5, 6, 7$ ; let this be $\vec{c}$ . Let $\vec{b}$ be the corresponding part of the $k$ -th row of $B$ . Again, loop through the index $k$ to add the resulting scalar-by-vector to the elements of $\vec{c}$ .
i = 1. Compute elements $C_{1j}$ for $j = 0, 1, 2, 3$ .
i = 1. Compute elements $C_{1j}$ for $j = 4, 5, 6, 7$ .

The “tiled” cache blocking SIMD approach performs slightly worse than the transposed SIMD approach.

C               0.768672 seconds
registers:      0.277462 seconds
simd,naive:     0.416584 seconds
simd,transpose: 0.275622 seconds
simd,tiled:     0.356344 seconds

5DGEMM 10: GCC Optimization¶

Finally, let’s compile using different gcc optimization flags. In Table 1, we can see that even with mild gcc optimizations like -O1, SIMD vastly outperforms any SISD approach.

Table 1:DGEMM SIMD runtime (in seconds) with different optimization flags.

Program	`-O0`	`-O1`	`-O2`	`-O3`
DGEMM (original)	0.768672	0.197168	0.197538	0.193889
DGEMM with registers	0.277462	0.191474	0.191921	0.126439
DGEMM, naive SIMD	0.416584	0.048405	0.049970	0.049045
DGEMM transpose SIMD	0.275622	0.031824	0.032188	0.031978
DGEMM tile SIMD	0.356344	0.033386	0.035177	0.037030

Generally speaking, most of the speedup comes not from doing multiple math operations at a time, but instead from doing a large memory load/store at a time.

6DGEMM SIMD Code¶

The three algorithsm discussed in this section leverage Intel SIMD extensions. You will see in the code below that the SIMD instructions are written in C as Intel Intrinsics. More next!

DGEMM, SIMD naive

DGEMM, SIMD Transpose

DGEMM, SIMD Tiled

DGEMM (original)

AVX intrinsics

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
matrix_d_t *dgemm_simd(matrix_d_t *A, matrix_d_t *B) {
  if (A->ncols!=B->nrows) return NULL; 
    matrix_d_t *C = init_mat_d(A->nrows, B->ncols);
  for (int i = 0; i < A->nrows; i++) {
    for (int j = 0; j < B->ncols; j+=4) { // 4 doubles at a time
            avx256_t v_C = avx_load(C->data + i*C->ncols +j);
            for (int k = 0; k < A->ncols; k++) {
                avx256_t s_A = avx_set_num(A->data[(i*A->ncols)+k]);
                avx256_t v_B = avx_load(B->data+k*B->ncols+j);
                // C_ij += a_ik * B_jk (for j = 0...3)
                v_C = avx_mul_add(s_A, v_B, v_C);
            }
            avx_store(C->data+i*C->ncols+j, v_C);
    }
  }
  return C;
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
matrix_d_t *dgemm_simd_transpose(matrix_d_t *A, matrix_d_t *B) {
  if (A->ncols!=B->nrows) return NULL;
    matrix_d_t *C = init_mat_d(A->nrows, B->ncols);
    matrix_d_t *B_T = transpose_mat_d(B);
  for (int i = 0; i < A->nrows; i++) {
    for (int j = 0; j < B->ncols; j++) {
      double *ptr_A = A->data+(i*A->ncols);
      double *ptr_B_T = B_T->data+(j*B_T->ncols);
      int k = 0;
  
            // 4 doubles at a time
            avx256_t v_C = avx_set_num(0);
            for (; k < A->ncols/4*4; k+= 4) {
                avx256_t v_A = avx_load(ptr_A+k);
                avx256_t v_B = avx_load(ptr_B_T+k);
                v_C = avx_mul_add(v_A, v_B, v_C);
            }
  
      double mem[4] __attribute__ ((aligned (64)));
            avx_store(mem, v_C);
            double sum = mem[0] + mem[1] + mem[2] + mem[3];

            // tail case
            for(; k < A->ncols; k++) {
                sum += ptr_A[k]*ptr_B_T[k];
            }
            C->data[i*C->ncols+j] = sum;
    }
  }
  free_mat_d(B_T);
  return C;
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
static inline void matmul_simd_tile(int si, int sj, int sk,
                 matrix_d_t *A, matrix_d_t *B, matrix_d_t *C) {
    for (int i = si; i < si + BLOCKSIZE; i++) {
        for (int j = sj; j < sj + BLOCKSIZE; j+=4) { // 4 doubles at a time
            avx256_t v_C = avx_load(C->data+i*C->ncols+j);

            for (int k = sk; k < sk + BLOCKSIZE; k++) {
                avx256_t s_A = avx_set_num(A->data[(i*A->ncols)+k]);
                avx256_t v_B = avx_load(B->data+k*B->ncols+j);
                v_C = avx_mul_add(s_A, v_B, v_C);
            }
            avx_store(C->data+i*C->ncols+j, v_C);
        }
    }
}
matrix_d_t *dgemm_simd_block(matrix_d_t *A, matrix_d_t *B) {
  if (A->ncols!=B->nrows) return NULL;
    matrix_d_t *C = init_mat_d(A->nrows, B->ncols);
  for (int si = 0; si < A->nrows; si += BLOCKSIZE) {
    for (int sj = 0; sj < B->ncols; sj += BLOCKSIZE) {
            for (int sk = 0; sk < A->ncols; sk+= BLOCKSIZE) {
                matmul_simd_tile(si, sj, sk, A, B, C);
            }
    }
  }
  return C;
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
matrix_d_t *dgemm(matrix_d_t *A, matrix_d_t *B) {
  if (A->ncols != B->nrows) return NULL;
  matrix_d_t *C = init_mat_d(A->nrows, B->ncols);
    
  for (int i = 0; i < A->nrows; i++) {
    for (int j = 0; j < B->ncols; j++) {
      double sum = 0; 
      for (int k = 0; k < A->ncols; k++) {
        sum += A->data[i*A->ncols+k]*B->data[k*B->ncols+j];
      }
      C->data[i*C->ncols+j] = sum;
    } 
  }   
  return C;
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
typedef __m256d avx256_t;

// Loads 8 doubles at memory address A into a avx256_t
static inline avx256_t avx_load(double *A) {
  return _mm256_load_pd(A);
}

// Stores the avx256_t at SRC to DST. Each avx256_t element gets stored in a 
// different index of the array passed into DST.
static inline void avx_store(double *dst, avx256_t src) {
  _mm256_store_pd(dst, src);
}

// Creates a avx256_t where every element is equal to num
static inline avx256_t avx_set_num(double num) {
  return _mm256_set1_pd(num);
}

// A * B + C
static inline avx256_t avx_mul_add(avx256_t A, avx256_t B, avx256_t C) {
  return _mm256_fmadd_pd(A, B, C);
}

Footnotes¶

In Figure 1, we assume a cache that has 256-bit blocks. This carries over the cache assumption from our cache blocking discussion.
↩