Skip to content

vulkan: fix im2col overflowing maxworkgroupcount #18180

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jeffbolznv wants to merge 1 commit into
base: master
Choose a base branch
from
+27 −6
Open

vulkan: fix im2col overflowing maxworkgroupcount #18180

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 9 additions & 2 deletions ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1258,6 +1258,7 @@ struct vk_op_im2col_push_constants {
int32_t s0; int32_t s1;
int32_t p0; int32_t p1;
int32_t d0; int32_t d1;
uint32_t batch_IC;
};

struct vk_op_im2col_3d_push_constants {
Expand Down Expand Up @@ -5870,6 +5871,9 @@ static void ggml_vk_dispatch_pipeline(ggml_backend_vk_context* ctx, vk_context&
std::cerr << "(" << buffer.buffer << ", " << buffer.offset << ", " << buffer.range << "), ";
}
std::cerr << "}, (" << wg0 << "," << wg1 << "," << wg2 << "))");
GGML_ASSERT(wg0 <= ctx->device->properties.limits.maxComputeWorkGroupCount[0] &&
wg1 <= ctx->device->properties.limits.maxComputeWorkGroupCount[1] &&
wg2 <= ctx->device->properties.limits.maxComputeWorkGroupCount[2]);
GGML_ASSERT(ctx->descriptor_set_idx < ctx->descriptor_sets.size());
GGML_ASSERT(descriptor_buffer_infos.size() <= MAX_PARAMETER_COUNT);
GGML_ASSERT(pipeline->parameter_count == descriptor_buffer_infos.size());
Expand Down Expand Up @@ -9056,6 +9060,8 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
const uint32_t batch = src1->ne[is_2D ? 3 : 2];

elements = { OW * KW * KH, OH, batch * IC };
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_IM2COL_3D:
{
Expand Down Expand Up @@ -10559,6 +10565,7 @@ static void ggml_vk_im2col(ggml_backend_vk_context * ctx, vk_context& subctx, co
const uint32_t batch_offset = src1->nb[is_2D ? 3 : 2] / 4; // nb is byte offset, src is type float32

const uint32_t pelements = OW * KW * KH;
const uint32_t batch = src1->ne[is_2D ? 3 : 2];

const ggml_backend_vk_buffer_context * d_buf_ctx = (ggml_backend_vk_buffer_context *)dst->buffer->context;
const vk_buffer d_buf = d_buf_ctx->dev_buffer;
Expand All @@ -10571,7 +10578,7 @@ static void ggml_vk_im2col(ggml_backend_vk_context * ctx, vk_context& subctx, co
IC, IW, IH, OW, OH, KW, KH,
pelements,
IC * KH * KW,
s0, s1, p0, p1, d0, d1,
s0, s1, p0, p1, d0, d1, batch * IC
});
}

Expand Down Expand Up @@ -14685,7 +14692,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
} else if (tensor->op == GGML_OP_LOG) {
tensor_clone = ggml_log(ggml_ctx, src_clone[0]);
} else if (tensor->op == GGML_OP_TRI) {
tensor_clone = ggml_tri(ggml_ctx, src_clone[0], ggml_get_op_params_i32(tensor, 0));
tensor_clone = ggml_tri(ggml_ctx, src_clone[0], (ggml_tri_type)ggml_get_op_params_i32(tensor, 0));
} else if (tensor->op == GGML_OP_DIAG) {
tensor_clone = ggml_diag(ggml_ctx, src_clone[0]);
} else if (tensor->op == GGML_OP_CLAMP) {
Expand Down
21 changes: 17 additions & 4 deletions ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@ layout (push_constant) uniform parameter
int s0; int s1;
int p0; int p1;
int d0; int d1;
uint batch_IC;
} p;

layout(constant_id = 0) const uint BLOCK_SIZE = 32;
Expand All @@ -34,12 +35,12 @@ layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
layout (buffer_reference) buffer D_ptr {D_TYPE d;};
#endif

void main() {
void im2col(const uint y, const uint z) {
const uint gidx = gl_GlobalInvocationID.x;

const uint oh = gl_GlobalInvocationID.y;
const uint batch = gl_GlobalInvocationID.z / p.IC;
const uint ic = gl_GlobalInvocationID.z % p.IC;
const uint oh = y;
const uint batch = z / p.IC;
const uint ic = z % p.IC;

const uint src_base = ic * p.offset_delta + batch * p.batch_offset;
const BDA_OFFSET_T dst_base = ((BDA_OFFSET_T(batch) * p.OH + oh) * p.OW) * p.CHW + BDA_OFFSET_T(ic) * (p.KW * p.KH);
Expand Down Expand Up @@ -101,3 +102,15 @@ void main() {
#endif
}
}

void main() {
uint y = gl_GlobalInvocationID.y;
while (y < p.OH) {
uint z = gl_GlobalInvocationID.z;
while (z < p.batch_IC) {
im2col(y, z);
z += gl_NumWorkGroups.z;
}
y += gl_NumWorkGroups.y;
}
}
1 change: 1 addition & 0 deletions tests/test-backend-ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6881,6 +6881,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {12, 12, 1, 2560}, {3, 3, 1, 2560}, 1, 1, 1, 1, 1, 1, true));
test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {12, 12, 2, 2560}, {3, 3, 2, 2560}, 1, 1, 1, 1, 1, 1, true));
test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_F16, {5, 5, 1, 32}, {3, 4, 1, 32}, 1, 1, 0, 0, 1, 1, true));
test_cases.emplace_back(new test_im2col(GGML_TYPE_F32, GGML_TYPE_F32, GGML_TYPE_F32, {2, 2, 1536, 729}, {2, 2, 1536, 4096}, 1, 1, 0, 0, 1, 1, true));

// im2col 3D
test_cases.emplace_back(new test_im2col_3d(GGML_TYPE_F32, GGML_TYPE_F32, GGML_TYPE_F32));
Expand Down
Loading