Skip to content

tests: hrx-backend-test expects the transposed V's spread pitch (a host check my #289 left stale; engine #1989) #294

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

Merged
bong-water-water-bong merged 2 commits into from
Nov 11, 2026
Merged

tests: hrx-backend-test expects the transposed V's spread pitch (a host check my #289 left stale; engine #1989) #294

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8288bc4
tests: hrx-backend-test compiles again after the alias_miss field (#262)
Nov 10, 2026
2136dd6
tests: hrx-backend-test expects the transposed V's spread pitch (engi…
Nov 10, 2026
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
15 changes: 11 additions & 4 deletions tests/hrx-backend-test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4119,8 +4119,15 @@ static void schedule_flash_attention_command(ggml_context * ctx,
REQUIRE(copy.bindings.size() == 2);
REQUIRE(copy.bindings[1].value == dispatch.bindings[2].value);
REQUIRE(copy.bindings[0].value != copy.bindings[1].value);
REQUIRE(copy.bindings[1].length == static_cast<size_t>(key_value_token_count * key_value_head_count * value_head_size) * sizeof(ggml_fp16_t));
require_compile_parameter(dispatch, "ggml.flash_attention.value_layout", "1");
// engine #1989: the transposed V has the spread pitch - keys + 32 when the key count is a multiple of 64 - which
// the transpose kernel is told and the attention kernel derives from its key count (value_layout 2). The source
// binding stays the packed V.
const int64_t transposed_pitch = key_value_token_count % 64 == 0 ? key_value_token_count + 32 : key_value_token_count;
REQUIRE(copy.bindings[0].length == static_cast<size_t>(key_value_token_count * key_value_head_count * value_head_size) * sizeof(ggml_fp16_t));
REQUIRE(copy.bindings[1].length == static_cast<size_t>(transposed_pitch * key_value_head_count * value_head_size) * sizeof(ggml_fp16_t));
require_compile_parameter(copy, "ggml.copy_transpose_f16.output_pitch", std::to_string(transposed_pitch));
require_compile_parameter(dispatch, "ggml.flash_attention.value_layout", "2");
REQUIRE(dispatch.kernel.compile_parameters.count("ggml.flash_attention.value_transposed_pitch") == 0);
} else {
REQUIRE(dispatch.kernel.compile_parameters.count("ggml.flash_attention.value_layout") == 0);
}
Expand Down Expand Up @@ -8375,15 +8382,15 @@ static void run_alias_value_import_checks() {
value_map
.add_snapshot_value({ ggml::hrx::ValueId(0), ggml::hrx::ValueKind::External, ggml::hrx::ValueStorageId(0),
ggml::hrx::ValueId(0), ggml::hrx::ValueId(), 0, 8 * sizeof(float), GGML_TYPE_F32, ne,
nb, 8, 8 * sizeof(float), true, nullptr, std::nullopt })
nb, 8, 8 * sizeof(float), true, nullptr, ggml::hrx::AliasMiss::None, std::nullopt })
.success());
REQUIRE(value_map.add_snapshot_storage({ ggml::hrx::ValueStorageId(1), ggml::hrx::ValueId(1), 8 * sizeof(float) })
.success());
REQUIRE(
value_map
.add_snapshot_value({ ggml::hrx::ValueId(1), ggml::hrx::ValueKind::Transient, ggml::hrx::ValueStorageId(1),
ggml::hrx::ValueId(1), ggml::hrx::ValueId(), 0, 8 * sizeof(float), GGML_TYPE_F32, ne,
nb, 8, 8 * sizeof(float), true, nullptr, std::nullopt })
nb, 8, 8 * sizeof(float), true, nullptr, ggml::hrx::AliasMiss::None, std::nullopt })
.success());
REQUIRE(value_map.alias_storage(ggml::hrx::ValueId(1), ggml::hrx::ValueId(0)).success());
const ggml::hrx::Value * aliased_value = value_map.find(ggml::hrx::ValueId(1));
Expand Down