Skip to content

tests: hrx-backend-test's two other stale host rows follow the backend (engine #2068) #295

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 3 commits into from
Nov 11, 2026
Merged

tests: hrx-backend-test's two other stale host rows follow the backend (engine #2068) #295

Changes from all commits
Commits
Show all changes
3 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
2334282
tests: hrx-backend-test's two other stale host rows follow the backen…
Nov 11, 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
25 changes: 18 additions & 7 deletions tests/hrx-backend-test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1835,8 +1835,11 @@ static void run_unmatched_graph_diagnostics_checks() {
ggml_context * ctx = ggml_init(params);
REQUIRE(ctx != nullptr);

ggml_tensor * cache = ggml_new_tensor_2d(ctx, GGML_TYPE_F16, 512, 512);
ggml_tensor * rows = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, 512, 2);
// A SET_ROWS that no registration takes. The i32-index form at an ordinary row width used to be that graph; since
// llama.cpp #196 the standalone hip_set_rows dispatch runs it on the device. A row wider than that dispatch's 32,768
// limit (and with i32 indices, which the two Loom writers decline) is still refused by every matcher.
ggml_tensor * cache = ggml_new_tensor_2d(ctx, GGML_TYPE_F16, 65536, 4);
ggml_tensor * rows = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, 65536, 2);
ggml_tensor * indices = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 2);
REQUIRE(cache != nullptr);
REQUIRE(rows != nullptr);
Expand Down Expand Up @@ -4119,8 +4122,15 @@ 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 @@ -6323,7 +6333,8 @@ static void run_qwen_matmul_dispatch_checks() {
REQUIRE(input != nullptr);
ggml_tensor * output = ggml_mul_mat(ctx, weight, input);
REQUIRE(output != nullptr);
REQUIRE(!matmul_graph_is_supported(ctx, output));
// llama.cpp #207: a batched dense MUL_MAT runs as one dispatch per plane; before it this graph was refused.
REQUIRE(matmul_graph_is_supported(ctx, output));
}

{
Expand Down Expand Up @@ -8375,15 +8386,15 @@ 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