Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-07 19:33:58 +01:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: Use push constant offset to handle misaligned descriptors (#10987)

Browse Source
This commit is contained in:
Jeff Bolz 2024-12-29 02:35:11 -06:00 committed by GitHub
parent f865ea149d
commit fdd2188912
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
19 changed files with 103 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

74
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -411,7 +411,7 @@ struct vk_op_unary_push_constants {
uint32_t ne;
uint32_t ne00; uint32_t ne01; uint32_t ne02; uint32_t ne03; uint32_t nb00; uint32_t nb01; uint32_t nb02; uint32_t nb03;
uint32_t ne10; uint32_t ne11; uint32_t ne12; uint32_t ne13; uint32_t nb10; uint32_t nb11; uint32_t nb12; uint32_t nb13;
uint32_t d_offset;
uint32_t misalign_offsets;
float param1; float param2;
uint32_t ne0_012mp; uint32_t ne0_012L;
uint32_t ne0_01mp; uint32_t ne0_01L;
@ -459,7 +459,7 @@ struct vk_op_binary_push_constants {
uint32_t ne00; uint32_t ne01; uint32_t ne02; uint32_t ne03; uint32_t nb00; uint32_t nb01; uint32_t nb02; uint32_t nb03;
uint32_t ne10; uint32_t ne11; uint32_t ne12; uint32_t ne13; uint32_t nb10; uint32_t nb11; uint32_t nb12; uint32_t nb13;
uint32_t ne20; uint32_t ne21; uint32_t ne22; uint32_t ne23; uint32_t nb20; uint32_t nb21; uint32_t nb22; uint32_t nb23;
uint32_t d_offset;
uint32_t misalign_offsets;
float param1; float param2; int32_t param3;
};
@ -546,7 +546,7 @@ struct vk_staging_memcpy {
};
struct vk_op_upscale_push_constants {
uint32_t ne; uint32_t d_offset;
uint32_t ne; uint32_t a_offset; uint32_t d_offset;
uint32_t nb00; uint32_t nb01; uint32_t nb02; uint32_t nb03;
uint32_t ne10; uint32_t ne11; uint32_t ne12; uint32_t ne13;
float sf0; float sf1; float sf2; float sf3;
@ -5076,6 +5076,57 @@ static bool ggml_vk_op_supports_incontiguous(ggml_op op) {
}
}
static uint32_t get_misalign_bytes(ggml_backend_vk_context * ctx, const ggml_tensor * t)
{
return ((vk_tensor_offset(t) + t->view_offs) & (ctx->device->properties.limits.minStorageBufferOffsetAlignment - 1));;
}
template <typename T> void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, T &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst) {
GGML_UNUSED(p);
GGML_UNUSED(src0);
GGML_UNUSED(src1);
GGML_UNUSED(src2);
GGML_UNUSED(dst);
static_assert(!std::is_const<T>::value, "unexpected type");
GGML_ASSERT(!src0 || get_misalign_bytes(ctx, src0) == 0);
GGML_ASSERT(!src1 || get_misalign_bytes(ctx, src1) == 0);
GGML_ASSERT(!src2 || get_misalign_bytes(ctx, src2) == 0);
GGML_ASSERT(!dst || get_misalign_bytes(ctx, dst) == 0);
}
template <> void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_unary_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst) {
const uint32_t a_offset = get_misalign_bytes(ctx, src0) / ggml_type_size(src0->type);
const uint32_t d_offset = get_misalign_bytes(ctx, dst) / ggml_type_size(dst->type);
p.misalign_offsets = (a_offset << 16) | d_offset;
GGML_UNUSED(src1);
GGML_UNUSED(src2);
}
template <> void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_binary_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst) {
const uint32_t a_offset = get_misalign_bytes(ctx, src0) / ggml_type_size(src0->type);
const uint32_t b_offset = get_misalign_bytes(ctx, src1) / ggml_type_size(src1->type);
const uint32_t d_offset = get_misalign_bytes(ctx, dst) / ggml_type_size(dst->type);
GGML_ASSERT(dst->op != GGML_OP_GET_ROWS || (a_offset == 0 && b_offset == 0 && d_offset == 0));
p.misalign_offsets = (a_offset << 16) | (b_offset << 8) | d_offset;
GGML_UNUSED(src2);
}
template <> void init_pushconst_tensor_offsets(ggml_backend_vk_context * ctx, vk_op_upscale_push_constants &p, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst) {
const uint32_t a_offset = get_misalign_bytes(ctx, src0) / ggml_type_size(src0->type);
const uint32_t d_offset = get_misalign_bytes(ctx, dst) / ggml_type_size(dst->type);
p.a_offset = a_offset;
p.d_offset = d_offset;
GGML_UNUSED(src1);
GGML_UNUSED(src2);
}
template<typename PC>
static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, const ggml_tensor * src2, ggml_tensor * dst, ggml_op op, PC&& pc, bool dryrun = false) {
VK_LOG_DEBUG("ggml_vk_op_f32((" << src0 << ", name=" << src0->name << ", type=" << src0->type << ", ne0=" << src0->ne[0] << ", ne1=" << src0->ne[1] << ", ne2=" << src0->ne[2] << ", ne3=" << src0->ne[3] << ", nb0=" << src0->nb[0] << ", nb1=" << src0->nb[1] << ", nb2=" << src0->nb[2] << ", nb3=" << src0->nb[3];
@ -5179,8 +5230,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
}
GGML_ASSERT(d_D != nullptr);
uint64_t d_buf_offset = ((vk_tensor_offset(dst) + dst->view_offs) / ctx->device->properties.limits.minStorageBufferOffsetAlignment) * ctx->device->properties.limits.minStorageBufferOffsetAlignment;
GGML_ASSERT(d_buf_offset == vk_tensor_offset(dst) || op == GGML_OP_CPY); // NOLINT
uint64_t d_buf_offset = vk_tensor_offset(dst) + dst->view_offs;
if(!src0_uma) {
d_X = src0_buf_ctx->dev_buffer;
x_buf_offset = vk_tensor_offset(src0) + src0->view_offs;
@ -5196,6 +5246,12 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
z_buf_offset = vk_tensor_offset(src2) + src2->view_offs;
GGML_ASSERT(d_Z != nullptr);
}
// Compute misalignment offset for descriptors and store it in in push constants, then align the descriptor offsets.
init_pushconst_tensor_offsets(ctx, pc, src0, src1, src2, dst);
x_buf_offset &= ~(ctx->device->properties.limits.minStorageBufferOffsetAlignment - 1);
y_buf_offset &= ~(ctx->device->properties.limits.minStorageBufferOffsetAlignment - 1);
z_buf_offset &= ~(ctx->device->properties.limits.minStorageBufferOffsetAlignment - 1);
d_buf_offset &= ~(ctx->device->properties.limits.minStorageBufferOffsetAlignment - 1);
if (op_supports_incontiguous) {
x_sz = ggml_nbytes(src0);
@ -5383,7 +5439,6 @@ static void ggml_vk_acc(ggml_backend_vk_context * ctx, vk_context& subctx, const
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t src1_type_size = ggml_type_size(src1->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
const uint32_t d_offset = ((vk_tensor_offset(dst) + dst->view_offs) % ctx->device->properties.limits.minStorageBufferOffsetAlignment) / dst_type_size;
int nb1 = dst->op_params[0] / 4; // 4 bytes of float32
int nb2 = dst->op_params[1] / 4; // 4 bytes of float32
@ -5395,7 +5450,7 @@ static void ggml_vk_acc(ggml_backend_vk_context * ctx, vk_context& subctx, const
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2],(uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)nb1, (uint32_t)nb2, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t)src1->ne[0], (uint32_t)src1->ne[1], (uint32_t)src1->ne[2],(uint32_t)src1->ne[3], (uint32_t)src1->nb[0] / src1_type_size, (uint32_t)src1->nb[1] / src1_type_size, (uint32_t)src1->nb[2] / src1_type_size, (uint32_t)src1->nb[3] / src1_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2],(uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t)nb1, (uint32_t)nb2, (uint32_t) dst->nb[3] / dst_type_size,
d_offset,
0,
0.0f, 0.0f, offset,
}, dryrun);
}
@ -5599,7 +5654,7 @@ static void ggml_vk_upscale(ggml_backend_vk_context * ctx, vk_context& subctx, c
const float sf3 = (float)dst->ne[3] / src0->ne[3];
ggml_vk_op_f32<vk_op_upscale_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_UPSCALE, {
(uint32_t)ggml_nelements(dst), 0,
(uint32_t)ggml_nelements(dst), 0, 0,
(uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t)dst->ne[0], (uint32_t)dst->ne[1], (uint32_t)dst->ne[2],(uint32_t)dst->ne[3],
sf0, sf1, sf2, sf3,
@ -5709,13 +5764,12 @@ static void ggml_vk_repeat(ggml_backend_vk_context * ctx, vk_context& subctx, co
static void ggml_vk_cpy(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
const uint32_t d_offset = ((vk_tensor_offset(dst) + dst->view_offs) % ctx->device->properties.limits.minStorageBufferOffsetAlignment) / dst_type_size;
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_CPY, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
d_offset,
0,
0.0f, 0.0f,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
}, dryrun);

4
ggml/src/ggml-vulkan/vulkan-shaders/acc.comp
View File

@ -21,9 +21,9 @@ void main() {
get_indices(idx, i00, i01, i02, i03);
if (ox < p.ne10 && oy < p.ne11 && oz < p.ne12) {
data_d[p.d_offset + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[ox + oy * p.ne10 + oz * p.ne10 * p.ne11]));
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[get_boffset() + ox + oy * p.ne10 + oz * p.ne10 * p.ne11]));
} else {
data_d[p.d_offset + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(i00, i01, i02, i03)]));
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]));
}
}

2
ggml/src/ggml-vulkan/vulkan-shaders/add.comp
View File

@ -22,7 +22,7 @@ void main() {
uint i00, i01, i02, i03;
get_indices(idx, i00, i01, i02, i03);
data_d[p.d_offset + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[src1_idx(i00, i01, i02, i03)]));
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) + FLOAT_TYPE(data_b[get_boffset() + src1_idx(i00, i01, i02, i03)]));
idx += num_threads;
}

4
ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp
View File

@ -12,6 +12,6 @@ void main() {
return;
}
const FLOAT_TYPE val = FLOAT_TYPE(data_a[src0_idx(idx)]);
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(val < p.param1 ? p.param1 : (val > p.param2 ? p.param2 : val));
const FLOAT_TYPE val = FLOAT_TYPE(data_a[get_aoffset() + src0_idx(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(val < p.param1 ? p.param1 : (val > p.param2 ? p.param2 : val));
}

6
ggml/src/ggml-vulkan/vulkan-shaders/concat.comp
View File

@ -30,12 +30,12 @@ void main() {
const bool is_src0 = i0 < p.ne00 && i1 < p.ne01 && i2 < p.ne02 && i3 < p.ne03;
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[p.d_offset + dst_idx] = D_TYPE(is_src0 ? data_a[src0_idx] : data_b[src1_idx]);
data_d[get_doffset() + dst_idx] = D_TYPE(is_src0 ? data_a[get_aoffset() + src0_idx] : data_b[get_boffset() + src1_idx]);
#else
if (is_src0) {
data_d[p.d_offset + dst_idx] = data_a[src0_idx];
data_d[get_doffset() + dst_idx] = data_a[get_aoffset() + src0_idx];
} else {
data_d[p.d_offset + dst_idx] = data_b[src1_idx];
data_d[get_doffset() + dst_idx] = data_b[get_boffset() + src1_idx];
}
#endif
}

8
ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp
View File

@ -19,9 +19,9 @@ void main() {
if (idx + (num_iter-1)*num_threads < p.ne) {
[[unroll]] for (uint i = 0; i < num_iter; ++i) {
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[p.d_offset + idx] = D_TYPE(data_a[idx]);
data_d[get_doffset() + idx] = D_TYPE(data_a[get_aoffset() + idx]);
#else
data_d[p.d_offset + idx] = data_a[idx];
data_d[get_doffset() + idx] = data_a[get_aoffset() + idx];
#endif
idx += num_threads;
}
@ -32,9 +32,9 @@ void main() {
}
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[p.d_offset + idx] = D_TYPE(data_a[idx]);
data_d[get_doffset() + idx] = D_TYPE(data_a[get_aoffset() + idx]);
#else
data_d[p.d_offset + idx] = data_a[idx];
data_d[get_doffset() + idx] = data_a[get_aoffset() + idx];
#endif
idx += num_threads;
}

4
ggml/src/ggml-vulkan/vulkan-shaders/copy.comp
View File

@ -13,8 +13,8 @@ void main() {
}
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(data_a[src0_idx(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(data_a[get_aoffset() + src0_idx(idx)]);
#else
data_d[p.d_offset + dst_idx(idx)] = data_a[src0_idx(idx)];
data_d[get_doffset() + dst_idx(idx)] = data_a[get_aoffset() + src0_idx(idx)];
#endif
}

4
ggml/src/ggml-vulkan/vulkan-shaders/cos.comp
View File

@ -12,6 +12,6 @@ void main() {
return;
}
const FLOAT_TYPE val = FLOAT_TYPE(data_a[src0_idx(idx)]);
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(cos(val));
const FLOAT_TYPE val = FLOAT_TYPE(data_a[get_aoffset() + src0_idx(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(cos(val));
}

2
ggml/src/ggml-vulkan/vulkan-shaders/div.comp
View File

@ -20,7 +20,7 @@ void main() {
uint i00, i01, i02, i03;
get_indices(idx, i00, i01, i02, i03);
data_d[p.d_offset + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(i00, i01, i02, i03)]) / FLOAT_TYPE(data_b[src1_idx(i00, i01, i02, i03)]));
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) / FLOAT_TYPE(data_b[get_boffset() + src1_idx(i00, i01, i02, i03)]));
idx += num_threads;
}

6
ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp
View File

@ -7,7 +7,7 @@ layout (push_constant) uniform parameter
uint ne00; uint ne01; uint ne02; uint ne03; uint nb00; uint nb01; uint nb02; uint nb03;
uint ne10; uint ne11; uint ne12; uint ne13; uint nb10; uint nb11; uint nb12; uint nb13;
uint ne20; uint ne21; uint ne22; uint ne23; uint nb20; uint nb21; uint nb22; uint nb23;
uint d_offset;
uint misalign_offsets;
float param1; float param2; int param3;
} p;
@ -22,6 +22,10 @@ uint get_idx() {
return gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
}
uint get_aoffset() { return p.misalign_offsets >> 16; }
uint get_boffset() { return (p.misalign_offsets >> 8) & 0xFF; }
uint get_doffset() { return p.misalign_offsets & 0xFF; }
// mod and div are expensive and coordinates/dimensions are often power of 2 or equal to 1
uint fastmod(uint a, uint b) {
if ((b & (b-1)) == 0) {

5
ggml/src/ggml-vulkan/vulkan-shaders/generic_unary_head.comp
View File

@ -6,7 +6,7 @@ layout (push_constant) uniform parameter
uint ne;
uint ne00; uint ne01; uint ne02; uint ne03; uint nb00; uint nb01; uint nb02; uint nb03;
uint ne10; uint ne11; uint ne12; uint ne13; uint nb10; uint nb11; uint nb12; uint nb13;
uint d_offset;
uint misalign_offsets;
float param1; float param2;
uint ne0_012mp; uint ne0_012L;
@ -24,6 +24,9 @@ uint get_idx() {
return gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
}
uint get_aoffset() { return p.misalign_offsets >> 16; }
uint get_doffset() { return p.misalign_offsets & 0xFFFF; }
// see init_fastdiv_values in ggml-vulkan.cpp
uint fastdiv(uint n, uint mp, uint L) {
uint msbs, lsbs;

6
ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp
View File

@ -15,10 +15,10 @@ void main() {
return;
}
const uint i01 = data_b[i10*p.nb10 + i11*p.nb11 + i12*p.nb12];
const uint i01 = data_b[get_boffset() + i10*p.nb10 + i11*p.nb11 + i12*p.nb12];
const uint a_offset = i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
const uint d_offset = i10*p.nb21 + i11*p.nb22 + i12*p.nb23;
const uint a_offset = get_aoffset() + i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
const uint d_offset = get_doffset() + i10*p.nb21 + i11*p.nb22 + i12*p.nb23;
#ifndef OPTIMIZATION_ERROR_WORKAROUND
data_d[d_offset + i00] = D_TYPE(data_a[a_offset + i00]);

2
ggml/src/ggml-vulkan/vulkan-shaders/mul.comp
View File

@ -20,7 +20,7 @@ void main() {
uint i00, i01, i02, i03;
get_indices(idx, i00, i01, i02, i03);
data_d[p.d_offset + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[src0_idx(i00, i01, i02, i03)]) * FLOAT_TYPE(data_b[src1_idx(i00, i01, i02, i03)]));
data_d[get_doffset() + dst_idx(i00, i01, i02, i03)] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + src0_idx(i00, i01, i02, i03)]) * FLOAT_TYPE(data_b[get_boffset() + src1_idx(i00, i01, i02, i03)]));
idx += num_threads;
}

2
ggml/src/ggml-vulkan/vulkan-shaders/pad.comp
View File

@ -24,5 +24,5 @@ void main() {
const bool is_src0 = i0 < p.ne00 && i1 < p.ne01 && i2 < p.ne02 && i3 < p.ne03;
data_d[p.d_offset + dst_idx] = D_TYPE(is_src0 ? data_a[src0_idx] : 0.0f);
data_d[get_doffset() + dst_idx] = D_TYPE(is_src0 ? data_a[get_aoffset() + src0_idx] : 0.0f);
}

2
ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp
View File

@ -22,5 +22,5 @@ void main() {
return;
}
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(data_a[src0_idx_mod(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(data_a[get_aoffset() + src0_idx_mod(idx)]);
}

2
ggml/src/ggml-vulkan/vulkan-shaders/scale.comp
View File

@ -18,7 +18,7 @@ void main() {
continue;
}
data_d[p.d_offset + idx] = D_TYPE(FLOAT_TYPE(data_a[idx]) * FLOAT_TYPE(p.param1));
data_d[get_doffset() + idx] = D_TYPE(FLOAT_TYPE(data_a[get_aoffset() + idx]) * FLOAT_TYPE(p.param1));
idx += num_threads;
}
}

4
ggml/src/ggml-vulkan/vulkan-shaders/sin.comp
View File

@ -12,6 +12,6 @@ void main() {
return;
}
const FLOAT_TYPE val = FLOAT_TYPE(data_a[src0_idx(idx)]);
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(sin(val));
const FLOAT_TYPE val = FLOAT_TYPE(data_a[get_aoffset() + src0_idx(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(sin(val));
}

4
ggml/src/ggml-vulkan/vulkan-shaders/square.comp
View File

@ -12,6 +12,6 @@ void main() {
return;
}
const FLOAT_TYPE val = FLOAT_TYPE(data_a[src0_idx(idx)]);
data_d[p.d_offset + dst_idx(idx)] = D_TYPE(val * val);
const FLOAT_TYPE val = FLOAT_TYPE(data_a[get_aoffset() + src0_idx(idx)]);
data_d[get_doffset() + dst_idx(idx)] = D_TYPE(val * val);
}

4
ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp
View File

@ -2,7 +2,7 @@
layout (push_constant) uniform parameter
{
uint ne; uint d_offset;
uint ne; uint a_offset; uint d_offset;
uint nb00; uint nb01; uint nb02; uint nb03;
uint ne10; uint ne11; uint ne12; uint ne13;
float sf0; float sf1; float sf2; float sf3;
@ -32,5 +32,5 @@ void main() {
const uint i02 = uint(i12 / p.sf2);
const uint i03 = uint(i13 / p.sf3);
data_d[p.d_offset + idx] = D_TYPE(data_a[i03 * p.nb03 + i02 * p.nb02 + i01 * p.nb01 + i00 * p.nb00]);
data_d[p.d_offset + idx] = D_TYPE(data_a[p.a_offset + i03 * p.nb03 + i02 * p.nb02 + i01 * p.nb01 + i00 * p.nb00]);
}