Skip to content

Commit

Permalink
Update CUDA ops ssm_conv and ssm_scan to match CPU implementation fro…
Browse files Browse the repository at this point in the history 
…m PR ggerganov#7531 (as per eb589d5)
  • Loading branch information
@jploski
jploski committed Jun 2, 2024
1 parent 9d0ccf8 commit 697fab6
Show file tree
Hide file tree
Showing 3 changed files with 136 additions and 220 deletions.
164 changes: 73 additions & 91 deletions ggml-cuda/ssm_conv.cu
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,13 +2,13 @@

template <int block_size>
static __global__ void ssm_conv_f32(
const float * src0, const float * src1, const float * src2, const float * src3,
const int src0_ne0, const int src0_nb1, const int src0_nb2,
const int src1_nb0, const int src1_nb1,
const int src2_nb1, const int src2_nb2,
const int src3_nb1,
const float * src0, const float * src1, const float * src2,
const int src0_nb1, const int src0_nb2,
const int src1_nb0, const int src1_nb1, const int src1_nb2,
const int src2_nb1,
float * dst,
const int nc, const int nr, const int n_t, const int n_kv) {
const int dst_nb0, const int dst_nb1, const int dst_nb2,
const int nc, const int nr, const int n_t, const int n_s) {

// const int row = blockIdx.x*blockDim.y + threadIdx.y;
const int tid = threadIdx.x;
Expand All @@ -24,136 +24,118 @@ static __global__ void ssm_conv_f32(
const int ir1 = min(ir0 + dr, nr);
const int ir = ir1 - ir0;

if (n_kv > 1) {
// multiple sequences means it's hard to know when it's the first time a state is read,
// so copy them all over to the destination, just to be sure.
for (int i3 = 0; i3 < n_kv; ++i3) {
float * s0 = (float *) ((char *) src0 + ir0*src0_nb1 + i3*src0_nb2);
float * s = (float *) ((char *) dst + ir0*src2_nb1 + i3*src2_nb2 + nr*n_t*sizeof(float));
// can't use memcpy because of d_conv vs d_conv - 1
for (int i1 = 0; i1 < ir; ++i1) {
for (int i0 = 0; i0 < nc - 1; ++i0) {
// copy s0 to last (d_conv - 1) columns of s
s[1 + i0 + i1*nc] = s0[i0 + i1*(nc - 1)];
}
}
}
}
// TODO: maybe require src0 to have d_conv columns instead of (d_conv - 1)?
// This would avoid having to copy into an intermediate buffer, but the state would be bigger.

for (int i2 = 0; i2 < n_t; ++i2) {
int32_t * sq = (int32_t *) ((char *) src3 + i2*src3_nb1); // {n_kv, n_tokens}
float * x = (float *) ((char *) dst + ir0*sizeof(float) + i2*(nr*sizeof(float))); // {d_inner, n_tokens}
float * s = (float *) ((char *) dst + ir0*src2_nb1 + sq[0]*src2_nb2 + nr*n_t*sizeof(float)); // {d_conv, d_inner, n_kv}
float * s0; // {d_conv - 1, d_inner, n_kv}
float * x0 = (float *) ((char *) src1 + ir0*src1_nb0 + i2*src1_nb1); // {d_inner, n_tokens}
float * c = (float *) ((char *) src2 + ir0*src2_nb1); // {d_conv, d_inner}
int ne0s0;

// avoid needing to copy the state for the first token
if (i2 == 0) {
s0 = (float *) ((char *) src0 + ir0*src0_nb1 + sq[0]*src0_nb2); // {d_conv - 1, d_inner, n_kv}
ne0s0 = src0_ne0;
} else {
// the source is the last (d_conv - 1) columns of the destination
s0 = s + 1;
ne0s0 = nc;
}
// float * s = (float *) params->wdata + (nc*dr + CACHE_LINE_SIZE_F32) * ith;
extern __shared__ float wdata_f32[]; // work buffer for all threads
float * s = (float *) wdata_f32 + nc*dr*ith;

// d_inner
for (int i3 = 0; i3 < n_s; ++i3) {
float * s0 = (float *) ((char *) src0 + ir0*src0_nb1) + i3*src0_nb2; // {d_conv, d_inner, n_s}

// copy the state into working memory
// can't use memcpy because (d_conv) != (d_conv - 1)
for (int i1 = 0; i1 < ir; ++i1) {
// shift state left
for (int i0 = 0; i0 < nc - 1; ++i0) {
s[i0 + i1*nc] = s0[i0 + i1*ne0s0];
s[1 + i0 + i1*nc] = s0[i0 + i1*(nc - 1)];
}
// insert x on the last column
s[(nc - 1) + i1*nc] = x0[i1];
}

// handle copies when there are multiple output states
for (int i3 = 1; i3 < n_kv; ++i3) {
int32_t seq = sq[i3];
if (0 <= seq && seq < n_kv) {
float * s1 = s + (seq - sq[0])*nc*nr;
for (int i2 = 0; i2 < n_t; ++i2) {
float * x = (float *) ((char *) dst + ir0* dst_nb0 + i2* dst_nb1 + i3* dst_nb2); // {d_inner, n_t, n_s}
float * x0 = (float *) ((char *) src1 + ir0*src1_nb0 + i2*src1_nb1 + i3*src1_nb2); // {d_inner, n_t, n_s}
float * c = (float *) ((char *) src2 + ir0*src2_nb1); // {d_conv, d_inner}

//memcpy(s1, s, nc*ir*sizeof(float));
for (int i4 = 0; i4 < nc*ir; i4++) {
s1[i4] = s[i4];
// shift state left
//memmove(s, s + 1, (nc*ir - 1) * sizeof(float));
for (int i4 = 0; i4 < nc*ir - 1; ++i4) {
s[i4] = s[i4+1];
}

// d_inner
for (int i1 = 0; i1 < ir; ++i1) {
// insert x on the last column
s[(nc - 1) + i1*nc] = x0[i1];
}

// it seems a little faster when this is separate from the state shift
for (int i1 = 0; i1 < ir; ++i1) {
// rowwise dot product
// NOTE: not using ggml_vec_dot_f32, because its sum is in double precision
float sumf = 0.0f;
for (int i0 = 0; i0 < nc; ++i0) {
int i = i0 + i1*nc;
sumf += s[i] * c[i];
}
} else {
// stop at negative or too big seq_ids
break;
x[i1] = sumf;
}
}

// it seems a little faster when this is separate from the state shift
// copy the state out of it
for (int i1 = 0; i1 < ir; ++i1) {
// rowwise dot product
float sumf = 0.0f;
for (int i0 = 0; i0 < nc; ++i0) {
int i = i0 + i1*nc;
sumf += s[i] * c[i];
for (int i0 = 0; i0 < nc - 1; ++i0) {
s0[i0 + i1*(nc - 1)] = s[1 + i0 + i1*nc];
}
x[i1] = sumf;
}
}
}

static void ssm_conv_f32_cuda(
const float * src0, const float * src1, const float * src2, const float * src3,
const int src0_ne0, const int src0_nb1, const int src0_nb2,
const int src1_nb0, const int src1_nb1,
const int src2_nb1, const int src2_nb2,
const int src3_nb1,
const float * src0, const float * src1, const float * src2,
const int src0_nb1, const int src0_nb2,
const int src1_nb0, const int src1_nb1, const int src1_nb2,
const int src2_nb1,
float * dst,
const int nc, const int nr, const int n_t, const int n_kv, cudaStream_t stream) {
const int dst_nb0, const int dst_nb1, const int dst_nb2,
const int nc, const int nr, const int n_t, const int n_s,
cudaStream_t stream) {

const dim3 block_dims(WARP_SIZE, 1, 1);
const int nblocks = 1; // TODO
const int shmem_size = nc * (nr + WARP_SIZE - 1) * sizeof(float); // TODO

ssm_conv_f32<WARP_SIZE><<<nblocks, block_dims, 0, stream>>>(
src0, src1, src2, src3,
src0_ne0, src0_nb1, src0_nb2,
src1_nb0, src1_nb1,
src2_nb1, src2_nb2,
src3_nb1,
ssm_conv_f32<WARP_SIZE><<<nblocks, block_dims, shmem_size, stream>>>(
src0, src1, src2,
src0_nb1, src0_nb2,
src1_nb0, src1_nb1, src1_nb2,
src2_nb1,
dst,
nc, nr, n_t, n_kv);
dst_nb0, dst_nb1, dst_nb2,
nc, nr, n_t, n_s);
}

void ggml_cuda_op_ssm_conv(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
const struct ggml_tensor * src0 = dst->src[0]; // conv_state
const struct ggml_tensor * src1 = dst->src[1]; // x
const struct ggml_tensor * src2 = dst->src[2]; // conv1d.weight
const struct ggml_tensor * src3 = dst->src[3]; // state_seq

const int nc = src2->ne[0]; // d_conv
const int nr = src0->ne[1]; // d_inner
const int n_t = src1->ne[1]; // n_tokens
const int n_kv = src0->ne[2]; // max number of sequences in the batch
const int nc = src2->ne[0]; // d_conv
const int nr = src0->ne[1]; // d_inner
const int n_t = src1->ne[1]; // tokens per sequence
const int n_s = src0->ne[2]; // number of sequences in the batch

GGML_ASSERT((nr*n_t) + (nc*nr*n_kv) == ggml_nelements(dst));
GGML_ASSERT(ggml_are_same_shape(src1, dst));
GGML_ASSERT(src0->nb[0] == sizeof(float));
GGML_ASSERT(src1->nb[0] == sizeof(float));
GGML_ASSERT(src2->nb[0] == sizeof(float));
GGML_ASSERT(src3->nb[0] == sizeof(int32_t));
GGML_ASSERT(src0->nb[1] == src0->ne[0]*sizeof(float));
// for use with the destination state offset between sequences
GGML_ASSERT(src2->nb[2] == src2->ne[1]*src2->ne[0]*sizeof(float));

const float * src0_d = (const float *)src0->data;
const float * src1_d = (const float *)src1->data;
const float * src2_d = (const float *)src2->data;
const float * src3_d = (const float *)src3->data;
float * dst_d = (float *)dst->data;
cudaStream_t stream = ctx.stream();

GGML_ASSERT(src0->type == GGML_TYPE_F32);
GGML_ASSERT( dst->type == GGML_TYPE_F32);

ssm_conv_f32_cuda(src0_d, src1_d, src2_d, src3_d,
src0->ne[0], src0->nb[1], src0->nb[2],
src1->nb[0], src1->nb[1],
src2->nb[1], src2->nb[2],
src3->nb[1],
dst_d, nc, nr, n_t, n_kv, stream);
ssm_conv_f32_cuda(src0_d, src1_d, src2_d,
src0->nb[1], src0->nb[2],
src1->nb[0], src1->nb[1], src1->nb[2],
src2->nb[1],
dst_d,
dst->nb[0], dst->nb[1], dst->nb[2],
nc, nr, n_t, n_s,
stream);
}
Loading

0 comments on commit 697fab6

Please sign in to comment.