Add f16 implementation to ggml_compute_forward_add_f16_f32
This commit is contained in:
parent
9ea91ceaf2
commit
19097c97a8
1 changed files with 37 additions and 14 deletions
51
ggml.c
51
ggml.c
|
@ -9358,9 +9358,15 @@ static void ggml_compute_forward_add_f16_f32(
|
|||
|
||||
GGML_ASSERT(src0->type == GGML_TYPE_F16);
|
||||
GGML_ASSERT(src1->type == GGML_TYPE_F32);
|
||||
GGML_ASSERT(dst->type == GGML_TYPE_F16);
|
||||
|
||||
GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
|
||||
if (dst->type == GGML_TYPE_F32) {
|
||||
GGML_ASSERT( nb0 == sizeof(float));
|
||||
}
|
||||
else {
|
||||
GGML_ASSERT(dst->type == GGML_TYPE_F16);
|
||||
GGML_ASSERT( nb0 == sizeof(ggml_fp16_t));
|
||||
}
|
||||
|
||||
GGML_ASSERT(nb00 == sizeof(ggml_fp16_t));
|
||||
|
||||
// rows per thread
|
||||
|
@ -9371,20 +9377,37 @@ static void ggml_compute_forward_add_f16_f32(
|
|||
const int ir1 = MIN(ir0 + dr, nr);
|
||||
|
||||
if (nb10 == sizeof(float)) {
|
||||
for (int ir = ir0; ir < ir1; ++ir) {
|
||||
// src0, src1 and dst are same shape => same indices
|
||||
const int i3 = ir/(ne2*ne1);
|
||||
const int i2 = (ir - i3*ne2*ne1)/ne1;
|
||||
const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
|
||||
if (dst->type == GGML_TYPE_F16) {
|
||||
for (int ir = ir0; ir < ir1; ++ir) {
|
||||
// src0, src1 and dst are same shape => same indices
|
||||
const int i3 = ir/(ne2*ne1);
|
||||
const int i2 = (ir - i3*ne2*ne1)/ne1;
|
||||
const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
|
||||
|
||||
ggml_fp16_t * dst_ptr = (ggml_fp16_t *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1);
|
||||
ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
|
||||
float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
|
||||
ggml_fp16_t * dst_ptr = (ggml_fp16_t *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1);
|
||||
ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
|
||||
float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
|
||||
|
||||
for (int i = 0; i < ne0; i++) {
|
||||
dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]);
|
||||
}
|
||||
}
|
||||
for (int i = 0; i < ne0; i++) {
|
||||
dst_ptr[i] = GGML_FP32_TO_FP16(GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i]);
|
||||
}
|
||||
}
|
||||
} else {
|
||||
for (int ir = ir0; ir < ir1; ++ir) {
|
||||
// src0, src1 and dst are same shape => same indices
|
||||
const int i3 = ir/(ne2*ne1);
|
||||
const int i2 = (ir - i3*ne2*ne1)/ne1;
|
||||
const int i1 = (ir - i3*ne2*ne1 - i2*ne1);
|
||||
|
||||
float * dst_ptr = (float *) ((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1);
|
||||
ggml_fp16_t * src0_ptr = (ggml_fp16_t *) ((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01);
|
||||
float * src1_ptr = (float *) ((char *) src1->data + i3*nb13 + i2*nb12 + i1*nb11);
|
||||
|
||||
for (int i = 0; i < ne0; i++) {
|
||||
dst_ptr[i] = GGML_FP16_TO_FP32(src0_ptr[i]) + src1_ptr[i];
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
// src1 is not contiguous
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue