vulkan: handle large sizes for get_rows (#15686)
This commit is contained in:
parent
c37052ab4d
commit
bbbf5ecccb
|
|
@ -7849,6 +7849,8 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
|
|||
break;
|
||||
case GGML_OP_GET_ROWS:
|
||||
elements = { (uint32_t)ne00, (uint32_t)ne10, (uint32_t)(ne11 * ne12) };
|
||||
elements[1] = std::min(elements[1], ctx->device->properties.limits.maxComputeWorkGroupCount[1]);
|
||||
elements[2] = std::min(elements[2], ctx->device->properties.limits.maxComputeWorkGroupCount[2]);
|
||||
break;
|
||||
case GGML_OP_ARGSORT:
|
||||
elements = { (uint32_t)ne00, (uint32_t)ggml_nrows(src0), 1 };
|
||||
|
|
|
|||
|
|
@ -7,14 +7,19 @@ layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
|||
|
||||
void main() {
|
||||
const uint i00 = gl_GlobalInvocationID.x;
|
||||
const uint i10 = gl_GlobalInvocationID.y;
|
||||
const uint i11 = (gl_GlobalInvocationID.z)/p.ne12;
|
||||
const uint i12 = (gl_GlobalInvocationID.z)%p.ne12;
|
||||
|
||||
if (i00 >= p.ne00) {
|
||||
return;
|
||||
}
|
||||
|
||||
uint gid_z = gl_GlobalInvocationID.z;
|
||||
while (gid_z < p.ne11 * p.ne12) {
|
||||
uint gid_y = gl_GlobalInvocationID.y;
|
||||
while (gid_y < p.ne10) {
|
||||
const uint i10 = gid_y;
|
||||
const uint i11 = gid_z / p.ne12;
|
||||
const uint i12 = gid_z % p.ne12;
|
||||
|
||||
const uint i01 = data_b[get_boffset() + i10*p.nb10 + i11*p.nb11 + i12*p.nb12];
|
||||
|
||||
const uint a_offset = get_aoffset() + i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
|
||||
|
|
@ -30,4 +35,8 @@ void main() {
|
|||
#else
|
||||
data_d[d_offset + i00] = D_TYPE(v);
|
||||
#endif
|
||||
gid_y += gl_WorkGroupSize.y * gl_NumWorkGroups.y;
|
||||
}
|
||||
gid_z += gl_WorkGroupSize.z * gl_NumWorkGroups.z;
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -10,9 +10,6 @@ layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
|||
|
||||
void main() {
|
||||
const uint i00 = (gl_GlobalInvocationID.x)*2;
|
||||
const uint i10 = gl_GlobalInvocationID.y;
|
||||
const uint i11 = (gl_GlobalInvocationID.z)/p.ne12;
|
||||
const uint i12 = (gl_GlobalInvocationID.z)%p.ne12;
|
||||
|
||||
#ifdef NEEDS_INIT_IQ_SHMEM
|
||||
init_iq_shmem(gl_WorkGroupSize);
|
||||
|
|
@ -22,6 +19,14 @@ void main() {
|
|||
return;
|
||||
}
|
||||
|
||||
uint gid_z = gl_GlobalInvocationID.z;
|
||||
while (gid_z < p.ne11 * p.ne12) {
|
||||
uint gid_y = gl_GlobalInvocationID.y;
|
||||
while (gid_y < p.ne10) {
|
||||
const uint i10 = gid_y;
|
||||
const uint i11 = gid_z / p.ne12;
|
||||
const uint i12 = gid_z % p.ne12;
|
||||
|
||||
const uint i01 = data_b[i10*p.nb10 + i11*p.nb11 + i12*p.nb12];
|
||||
|
||||
const uint a_offset = i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
|
||||
|
|
@ -38,4 +43,9 @@ void main() {
|
|||
|
||||
data_d[d_offset + iybs + iqs ] = D_TYPE(v.x);
|
||||
data_d[d_offset + iybs + iqs + y_offset] = D_TYPE(v.y);
|
||||
|
||||
gid_y += gl_WorkGroupSize.y * gl_NumWorkGroups.y;
|
||||
}
|
||||
gid_z += gl_WorkGroupSize.z * gl_NumWorkGroups.z;
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in New Issue