Skip to content

strided_copy: fix the num_aie_channels>1 device hang, and add test coverage - #158

Merged
andrej merged 4 commits into
amd:develfrom
atassis:pr/strided-copy-transfer-size
Aug 29, 2026
Merged

andrej merged 4 commits into
amd:develfrom
atassis:pr/strided-copy-transfer-size

Conversation

@atassis

@atassis atassis commented Aug 26, 2026

Copy link
Copy Markdown
Contributor

Stacked on #157: the new tests gate at exact equality, which #157 is what makes
expressible. Its two commits are in this diff and drop out when it merges.

transfer_size defaults to the product of the whole input_sizes, but the channel
split happens afterwards, so at num_aie_channels > 1 the ObjectFifo object is N times
the length of the shim BD feeding it. A BD shorter than its object starves the MemTile's
S2MM: it never completes an object, never releases the lock, the linked MM2S never
starts, and the drain's dma_await_task never returns. The host sees
ERT_CMD_STATE_TIMEOUT. Generation, the MLIR verifier and the xclbin build are all clean.

An integer multiple is fine — it just cycles the buffer. Stated as that ratio, it
predicted 8 of 8 arms on a sweep across channels 1/2/4 x transfer_size. num_aie_channels=4
lands on both sides — passes at transfer_size=256, hangs at 512 — which is what rules
out a multi-channel defect and leaves the ratio.

The operator also had no tests, and could not have had: get_arg_spec passed a bare int
where AIERuntimeArgSpec.shape is a tuple, and XRTTensor only treats a tuple as a shape.
Anything else goes through np.asarray, so the int becomes a 0-d array and the output
buffer is allocated for one element. It is the only one of the tree's 35 arg-spec call
sites that passes a scalar.

Added

  • iron/operators/strided_copy/test.py and reference.py. Arms cover contiguous, chunked,
    2 and 4 channels, and Llama's KV-cache write shape at several slots, plus a construction-time
    rejection test. Gated at rel_tol=abs_tol=0 — the operator does no arithmetic.

Changed

  • design.py: size the object against the per-channel share, and assert the divisibility
    the hardware requires, so ratio < 1 is a construction-time rejection instead of a device
    hang. Also assert that a copy moves the same element count both ways, which nothing
    checked and which fails the same undiagnosable way on the output side.
  • op.py: the arg spec now passes a shape tuple.

Removed

Evidence

On devel with design.py reverted, exactly the four multi-channel arms fail with
ERT_CMD_STATE_TIMEOUT and the single-channel and chunked arms pass. With the fix,
11/11 pass at exact equality. Arithmetic at num_aie_channels=1 is unchanged, so
llama_npu.py's KV-cache write — the only in-tree caller — is untouched.

Known gap

llama_npu.py drives the output offset at runtime through a ScratchpadParameter
(output_offset_parameter="cache_offset"). The new arms bake the offset as a compile-time
constant instead, so that path is still uncovered.

@andrej andrej left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good fix, and thanks for adding tests to an operator that previously had none.

Comment thread iron/operators/strided_copy/test.py Outdated
Comment thread iron/operators/strided_copy/test.py Outdated
atassis and others added 3 commits August 28, 2026 16:31
AIERuntimeArgSpec.shape is a tuple and run_test builds the output XRTTensor
straight from it. XRTTensor only treats a tuple as a shape; anything else goes
through np.asarray, so a bare int becomes a 0-d array and the buffer is
allocated for one element. Every comparison then fails on a size mismatch
before the operator's own behaviour is reached. It is the only one of the
tree's 35 arg-spec call sites that passes a scalar.
transfer_size defaulted to the product of the whole input_sizes, but the channel
split happens afterwards, so at num_aie_channels>1 the object was N times the
length of the BD feeding it. A BD shorter than its object starves the MemTile's
S2MM: it never completes an object, never releases the lock, and the drain's
dma_await_task never returns (ERT_CMD_STATE_TIMEOUT), with nothing reported by
generation or the xclbin build. An integer multiple is fine, it just cycles the
buffer. Default to the per-channel share and assert the divisibility, so a
ratio below 1 is a construction-time rejection instead of a device hang.
Arithmetic at num_aie_channels=1 is unchanged.

Also assert that a copy moves the same element count both ways, which nothing
checked and which fails the same undiagnosable way on the output side.

Adds the operator's first test coverage, plus a reference that models the taps.
Co-authored-by: André Rösti <androsti@amd.com>
@andrej
andrej force-pushed the pr/strided-copy-transfer-size branch from 3a5dabb to ab937ac Compare August 28, 2026 22:32
@andrej
andrej enabled auto-merge August 28, 2026 22:58
@github-actions

github-actions Bot commented Aug 28, 2026

Copy link
Copy Markdown
Contributor

CI Test Results

6ea8fb8 (2026_08_29_03_27_57)

IRON - CI Summary

Examples

iron/applications/llama_3.2_1b
Test Krackan Status Krackan Phoenix Status Phoenix
test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_1] - - -
test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_40] - - -
test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_1] - - -
test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_40] - - -

Small

iron/operators/axpy
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0] - - 425.94
test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0] - - 382.60
test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0] - - 438.58
iron/operators/dequant
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32] - - 320.76
test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32] - - 348.60
test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32] - - 365.92
test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32] - - 427.98
test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32] - - 532.90
test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32] - - 458.54
iron/operators/elementwise_add
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048] - - 394.12
test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024] - - 464.38
test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512] - - 368.98
iron/operators/elementwise_mul
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048] - - 441.72
test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024] - - 401.82
test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512] - - 406.30
iron/operators/gelu
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 307.36
test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] - - 429.84
test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 358.00
test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] - - 361.74
test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 761.46
test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] - - 900.84
iron/operators/gemm
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1] - - 524.30
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1] - - 532.58
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1] - - 82379.64
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1] - - 25151.50
test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1] - - 3365.68
test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4] - - 6171.84
iron/operators/gemv
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128] - - 0.12
test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048] - - 3.73
test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024] - - 6.21
test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512] - - 10.27
test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024] - - 3.60
test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024] - - 5.83
test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024] - - 8.40
test_gemv_batched[M_1024-K_1024-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_2] - - 2.11
test_gemv_batched[M_1026-K_64-num_aie_columns_1-tile_size_input_1-tile_size_output_2-num_batches_2] - - 0.38
test_gemv_batched[M_256-K_128-num_aie_columns_1-tile_size_input_1-tile_size_output_256-num_batches_4] - - 0.30
test_gemv_batched[M_64-K_1536-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_8] - - 1.20
test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128] - - -
test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048] - - -
test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024] - - -
iron/operators/layer_norm
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 380.64
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] - - 340.98
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 366.86
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] - - 353.30
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 430.40
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] - - 453.20
iron/operators/leaky_relu
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01] - - 607.02
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1] - - 334.60
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25] - - 388.66
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01] - - 306.28
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01] - - 435.12
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01] - - 475.94
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01] - - 393.66
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01] - - 529.32
iron/operators/mem_copy
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048] - - 417.18
test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024] - - 450.52
test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024] - - 342.98
test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512] - - 472.18
test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512] - - 457.78
test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256] - - 484.62
iron/operators/relu
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 385.40
test_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] - - 366.16
test_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 378.18
test_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] - - 316.50
test_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 373.50
test_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] - - 433.62
iron/operators/rms_norm
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False] - - 414.44
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True] - - 313.14
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False] - - 480.56
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True] - - 307.56
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False] - - 387.22
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True] - - 381.34
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False] - - 362.04
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True] - - 576.46
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False] - - 467.30
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True] - - 437.06
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False] - - 479.36
iron/operators/rope
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0] - - 473.40
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0] - - 394.24
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0] - - 369.94
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0] - - 303.78
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0] - - 401.60
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0] - - 374.90
iron/operators/sigmoid
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 351.14
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] - - 364.28
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 439.52
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] - - 513.10
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 415.32
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] - - 801.30
iron/operators/silu
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_silu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 370.44
test_silu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 463.52
test_silu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 758.92
iron/operators/softmax
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024] - - 422.22
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048] - - 475.54
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512] - - 429.60
iron/operators/strided_copy
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_strided_copy[chunked_transfer] - - 476.50
test_strided_copy[contiguous] - - 694.26
test_strided_copy[four_channels] - - 264.26
test_strided_copy[kv_slot0] - - 262.80
test_strided_copy[kv_slot5] - - 329.76
test_strided_copy[kv_slot5_four_channels] - - 380.42
test_strided_copy[kv_slot5_two_channels] - - 318.58
test_strided_copy[kv_slot_last] - - 325.40
test_strided_copy[two_channels] - - 300.38
test_strided_copy[two_channels_chunked] - - 356.90
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter0] - - -
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter1] - - -
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter2] - - -
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter3] - - -
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter4] - - -
iron/operators/swiglu_decode
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_swiglu_decode[embedding_dim_1024-hidden_dim_3584] - - 8839.95
test_swiglu_decode[embedding_dim_2048-hidden_dim_2048] - - 15665.92
iron/operators/swiglu_prefill
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False] - - 24198.95
iron/operators/tanh
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_tanh[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048] - - 391.66
test_tanh[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024] - - 431.66
test_tanh[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024] - - 402.12
test_tanh[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512] - - 449.24
test_tanh[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512] - - 418.36
test_tanh[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256] - - 797.90
iron/operators/transpose
Test Krackan Status Krackan Phoenix Status Phoenix Latency (mean)
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1] - - 494.38
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2] - - 625.86
test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1] - - 816.64
Krackan - Examples

IRON

Tested on 2026_08_29_03_27_57 at commit 6ea8fb8.

iron/applications/llama_3.2_1b
TestChecksTTFT (mean)TPS (mean)
test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_1]✅ 5/52.07n/a
test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_40]✅ 5/52.097.80
test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_1]✅ 5/52.03n/a
test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_40]✅ 5/52.037.58

Trends:

IRON Trends

iron/applications/llama_3.2_1b

test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_1]

Commit/Date TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
6ea8fb8 — 2026-08-29 03:22:402.08 (+0.24%)2.07 (+0.55%)2.08 (+1.42%)2.05 (+0.29%)0.02 (+11.10%)
3ad2eb2 — 2026-08-29 00:45:192.08 (n/a)2.05 (n/a)2.05 (n/a)2.04 (n/a)0.02 (n/a)

test_llama_3_2_1b[llama_3.2_1b_prompt_1024_tokens_40]

Commit/Date TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
6ea8fb8 — 2026-08-29 03:22:407.81 (-0.14%)7.80 (-0.13%)7.80 (-0.09%)7.79 (-0.17%)0.01 (+18.61%)2.22 (+2.30%)2.09 (+0.57%)2.06 (-0.15%)2.05 (+0.54%)0.07 (+38.65%)
3ad2eb2 — 2026-08-29 00:45:197.83 (n/a)7.81 (n/a)7.81 (n/a)7.80 (n/a)0.01 (n/a)2.17 (n/a)2.08 (n/a)2.06 (n/a)2.04 (n/a)0.05 (n/a)

test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_1]

Commit/Date TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
6ea8fb8 — 2026-08-29 03:22:402.04 (-0.59%)2.03 (+0.27%)2.03 (+0.59%)2.02 (+0.90%)0.01 (-58.50%)
3ad2eb2 — 2026-08-29 00:45:192.05 (n/a)2.02 (n/a)2.02 (n/a)2.00 (n/a)0.02 (n/a)

test_llama_3_2_1b[llama_3.2_1b_prompt_13_tokens_40]

Commit/Date TPS (max)TPS (mean)TPS (median)TPS (min)TPS (stddev)TTFT (max)TTFT (mean)TTFT (median)TTFT (min)TTFT (stddev)
6ea8fb8 — 2026-08-29 03:22:407.61 (+0.04%)7.58 (+0.01%)7.60 (+0.21%)7.55 (-0.11%)0.02 (+34.35%)2.06 (+0.59%)2.03 (-0.14%)2.03 (-0.10%)1.98 (-1.35%)0.03 (+83.18%)
3ad2eb2 — 2026-08-29 00:45:197.60 (n/a)7.58 (n/a)7.58 (n/a)7.56 (n/a)0.02 (n/a)2.05 (n/a)2.03 (n/a)2.04 (n/a)2.01 (n/a)0.02 (n/a)
Phoenix - Small

IRON

Tested on 2026_08_29_03_27_15 at commit 6ea8fb8.

iron/operators/axpy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]✅ 5/5425.940.03n/a
test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]✅ 5/5382.600.04n/a
test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]✅ 5/5438.580.03n/a
iron/operators/dequant
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]✅ 5/5320.760.02n/a
test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]✅ 5/5348.600.02n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]✅ 5/5365.920.02n/a
test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]✅ 5/5427.980.01n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]✅ 5/5532.900.01n/a
test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]✅ 5/5458.540.01n/a
iron/operators/elementwise_add
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5394.120.03n/a
test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5464.380.03n/a
test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5368.980.03n/a
iron/operators/elementwise_mul
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]✅ 5/5441.720.03n/a
test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]✅ 5/5401.820.04n/a
test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]✅ 5/5406.300.03n/a
iron/operators/gelu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5307.360.03n/a
test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5429.840.02n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5358.000.03n/a
test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5361.740.02n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5761.460.02n/a
test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5900.840.01n/a
iron/operators/gemm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5524.300.4619.50
test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]✅ 5/5532.580.4418.72
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/582379.640.31208.57
test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]✅ 5/525151.501.00683.37
test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]✅ 5/53365.682.63690.20
test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]✅ 5/56171.840.2111.43
iron/operators/gemv
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]✅ 5/5n/a0.120.11
test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]✅ 5/5n/a3.733.73
test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]✅ 5/5n/a6.216.20
test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]✅ 5/5n/a10.2710.27
test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a3.603.60
test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a5.835.83
test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]✅ 5/5n/a8.408.39
test_gemv_batched[M_1024-K_1024-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_2]✅ 5/5n/a2.112.11
test_gemv_batched[M_1026-K_64-num_aie_columns_1-tile_size_input_1-tile_size_output_2-num_batches_2]✅ 5/5n/a0.380.38
test_gemv_batched[M_256-K_128-num_aie_columns_1-tile_size_input_1-tile_size_output_256-num_batches_4]✅ 5/5n/a0.300.30
test_gemv_batched[M_64-K_1536-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_8]✅ 5/5n/a1.201.18
test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]❌ 0/5n/an/an/a
test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]❌ 0/5n/an/an/a
test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]❌ 0/5n/an/an/a
iron/operators/layer_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5380.640.02n/a
test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5340.980.03n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5366.860.03n/a
test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5353.300.02n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5430.400.02n/a
test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5453.200.02n/a
iron/operators/leaky_relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]✅ 5/5607.020.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]✅ 5/5334.600.03n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]✅ 5/5388.660.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]✅ 5/5306.280.03n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]✅ 5/5435.120.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]✅ 5/5475.940.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]✅ 5/5393.660.02n/a
test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]✅ 5/5529.320.02n/a
iron/operators/mem_copy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]✅ 5/5417.180.02n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]✅ 5/5450.520.02n/a
test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]✅ 5/5342.980.02n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]✅ 5/5472.180.02n/a
test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]✅ 5/5457.780.02n/a
test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]✅ 5/5484.620.02n/a
iron/operators/relu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5385.400.02n/a
test_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5366.160.03n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5378.180.03n/a
test_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5316.500.03n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5373.500.02n/a
test_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5433.620.02n/a
iron/operators/rms_norm
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]✅ 5/5414.440.02n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]✅ 5/5313.140.04n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]✅ 5/5480.560.02n/a
test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]✅ 5/5307.560.04n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]✅ 5/5387.220.02n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]✅ 5/5381.340.03n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]✅ 5/5362.040.02n/a
test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]✅ 5/5576.460.02n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]✅ 5/5467.300.02n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]✅ 5/5437.060.02n/a
test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]✅ 5/5479.360.02n/a
iron/operators/rope
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]✅ 5/5473.400.25n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]✅ 5/5394.240.29n/a
test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]✅ 5/5369.940.30n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]✅ 5/5303.780.26n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]✅ 5/5401.600.21n/a
test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]✅ 5/5374.900.24n/a
iron/operators/sigmoid
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5351.140.02n/a
test_sigmoid[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5364.280.03n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5439.520.02n/a
test_sigmoid[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5513.100.02n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5415.320.02n/a
test_sigmoid[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5801.300.01n/a
iron/operators/silu
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_silu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5370.440.02n/a
test_silu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5463.520.02n/a
test_silu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5758.920.02n/a
iron/operators/softmax
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]✅ 5/5422.220.34n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]✅ 5/5475.540.31n/a
test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5429.600.35n/a
iron/operators/strided_copy
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_strided_copy[chunked_transfer]✅ 5/5476.500.01n/a
test_strided_copy[contiguous]✅ 5/5694.260.01n/a
test_strided_copy[four_channels]✅ 5/5264.260.02n/a
test_strided_copy[kv_slot0]✅ 5/5262.800.51n/a
test_strided_copy[kv_slot5]✅ 5/5329.760.46n/a
test_strided_copy[kv_slot5_four_channels]✅ 5/5380.420.37n/a
test_strided_copy[kv_slot5_two_channels]✅ 5/5318.580.43n/a
test_strided_copy[kv_slot_last]✅ 5/5325.400.43n/a
test_strided_copy[two_channels]✅ 5/5300.380.01n/a
test_strided_copy[two_channels_chunked]✅ 5/5356.900.01n/a
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter0]✅ 1/1n/an/an/a
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter1]✅ 1/1n/an/an/a
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter2]✅ 1/1n/an/an/a
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter3]✅ 1/1n/an/an/a
test_transfer_size_not_dividing_per_channel_share_is_rejected[iter4]✅ 1/1n/an/an/a
iron/operators/swiglu_decode
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]✅ 5/58839.950.00n/a
test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]✅ 5/515665.920.00n/a
iron/operators/swiglu_prefill
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]✅ 5/524198.950.09n/a
iron/operators/tanh
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_tanh[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]✅ 5/5391.660.02n/a
test_tanh[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]✅ 5/5431.660.02n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]✅ 5/5402.120.02n/a
test_tanh[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]✅ 5/5449.240.02n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]✅ 5/5418.360.02n/a
test_tanh[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]✅ 5/5797.900.01n/a
iron/operators/transpose
TestChecksLatency (mean)Bandwidth (mean)Throughput (mean)
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]✅ 5/5494.381.30n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]✅ 5/5625.861.75n/a
test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]✅ 5/5816.640.92n/a

Trends:

IRON Trends

iron/operators/axpy

test_axpy[input_length_2048-num_aie_columns_1-tile_size_2048-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (+2.37%)0.03 (-14.58%)0.03 (-34.77%)0.02 (-1.54%)0.01 (+0.74%)528.20 (+1.56%)425.94 (+17.29%)460.00 (+53.28%)275.70 (-2.30%)107.73 (+3.15%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.01 (n/a)520.10 (n/a)363.14 (n/a)300.10 (n/a)282.20 (n/a)104.44 (n/a)

test_axpy[input_length_2048-num_aie_columns_2-tile_size_1024-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (+3.21%)0.04 (-19.49%)0.04 (-19.79%)0.02 (-19.56%)0.01 (+21.63%)599.40 (+24.31%)382.60 (+31.47%)294.80 (+24.65%)224.10 (-3.11%)160.95 (+49.15%)
3ad2eb2 — 2026-08-29 00:27:500.05 (n/a)0.05 (n/a)0.05 (n/a)0.03 (n/a)0.01 (n/a)482.20 (n/a)291.02 (n/a)236.50 (n/a)231.30 (n/a)107.91 (n/a)

test_axpy[input_length_2048-num_aie_columns_4-tile_size_512-scalar_factor_3.0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (-29.69%)0.03 (-22.22%)0.03 (-33.44%)0.02 (-10.47%)0.01 (-35.07%)604.70 (+11.69%)438.58 (+22.64%)473.80 (+50.27%)292.70 (+42.23%)136.94 (-5.01%)
3ad2eb2 — 2026-08-29 00:27:500.06 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.02 (n/a)541.40 (n/a)357.62 (n/a)315.30 (n/a)205.80 (n/a)144.16 (n/a)
iron/operators/dequant

test_dequant[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (-13.28%)0.02 (+20.80%)0.02 (+45.34%)0.01 (+10.10%)0.00 (-35.77%)498.10 (-9.17%)320.76 (-22.60%)293.50 (-31.20%)220.90 (+15.29%)104.21 (-22.85%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)548.40 (n/a)414.42 (n/a)426.60 (n/a)191.60 (n/a)135.07 (n/a)

test_dequant[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (+25.15%)0.02 (+12.32%)0.02 (+52.44%)0.01 (+4.37%)0.01 (+11.42%)513.90 (-4.19%)348.60 (-11.20%)312.50 (-34.39%)181.70 (-20.10%)137.34 (-8.09%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)536.40 (n/a)392.56 (n/a)476.30 (n/a)227.40 (n/a)149.43 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (+4.01%)0.02 (+5.67%)0.01 (+15.04%)0.01 (+1.17%)0.01 (+7.96%)498.80 (-1.15%)365.92 (-4.24%)374.00 (-13.06%)217.40 (-3.89%)134.65 (+3.85%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)504.60 (n/a)382.14 (n/a)430.20 (n/a)226.20 (n/a)129.66 (n/a)

test_dequant[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (-13.22%)0.01 (-28.17%)0.01 (-32.97%)0.01 (-31.51%)0.00 (+21.72%)533.90 (+45.99%)427.98 (+44.37%)450.10 (+49.19%)258.40 (+15.25%)104.53 (+96.95%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)365.70 (n/a)296.44 (n/a)301.70 (n/a)224.20 (n/a)53.07 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (+12.95%)0.01 (-0.15%)0.01 (+10.81%)0.01 (+142.34%)0.01 (-3.29%)765.30 (-58.74%)532.90 (-24.10%)480.60 (-9.76%)215.40 (-11.47%)230.54 (-65.10%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)0.01 (n/a)1854.70 (n/a)702.10 (n/a)532.60 (n/a)243.30 (n/a)660.50 (n/a)

test_dequant[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-group_size_32]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (+65.62%)0.01 (+22.72%)0.01 (-4.37%)0.01 (+57.14%)0.01 (+78.76%)554.30 (-36.36%)458.54 (-17.07%)512.00 (+4.55%)225.50 (-39.63%)133.09 (-34.42%)
3ad2eb2 — 2026-08-29 00:27:500.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)871.00 (n/a)552.90 (n/a)489.70 (n/a)373.50 (n/a)202.95 (n/a)
iron/operators/elementwise_add

test_elementwise_add[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (n/a)0.03 (n/a)0.04 (n/a)0.02 (n/a)0.01 (n/a)578.10 (n/a)394.12 (n/a)331.00 (n/a)265.40 (n/a)140.31 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (n/a)0.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)577.30 (n/a)464.38 (n/a)505.70 (n/a)267.70 (n/a)128.63 (n/a)

test_elementwise_add[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (n/a)0.03 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)471.80 (n/a)368.98 (n/a)353.00 (n/a)279.50 (n/a)85.74 (n/a)
iron/operators/elementwise_mul

test_elementwise_mul[input_length_2048-num_aie_columns_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)612.10 (n/a)441.72 (n/a)452.30 (n/a)267.20 (n/a)123.01 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.06 (n/a)0.04 (n/a)0.03 (n/a)0.02 (n/a)0.02 (n/a)557.80 (n/a)401.82 (n/a)468.90 (n/a)198.40 (n/a)155.72 (n/a)

test_elementwise_mul[input_length_2048-num_aie_columns_4-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)550.00 (n/a)406.30 (n/a)437.90 (n/a)228.20 (n/a)121.35 (n/a)
iron/operators/gelu

test_gelu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)464.90 (n/a)307.36 (n/a)293.40 (n/a)237.40 (n/a)92.12 (n/a)

test_gelu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)583.40 (n/a)429.84 (n/a)514.70 (n/a)220.20 (n/a)155.70 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)508.60 (n/a)358.00 (n/a)300.10 (n/a)221.30 (n/a)124.13 (n/a)

test_gelu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.02 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)600.20 (n/a)361.74 (n/a)297.40 (n/a)255.80 (n/a)139.74 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.02 (n/a)0.02 (n/a)0.00 (n/a)0.01 (n/a)1943.50 (n/a)761.46 (n/a)514.30 (n/a)271.90 (n/a)671.74 (n/a)

test_gelu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)0.01 (n/a)2160.30 (n/a)900.84 (n/a)662.00 (n/a)374.00 (n/a)714.87 (n/a)
iron/operators/gemm

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_False-c_col_maj_False-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.61 (-12.50%)0.46 (+0.63%)0.44 (+18.97%)0.31 (-8.95%)0.14 (-6.38%)717.20 (+9.83%)524.30 (-0.11%)501.50 (-15.94%)363.40 (+14.28%)164.23 (+14.02%)25.97 (-12.50%)19.50 (+0.63%)18.82 (+18.97%)13.16 (-8.95%)6.06 (-6.38%)
3ad2eb2 — 2026-08-29 00:27:500.70 (n/a)0.45 (n/a)0.37 (n/a)0.34 (n/a)0.15 (n/a)653.00 (n/a)524.90 (n/a)596.60 (n/a)318.00 (n/a)144.04 (n/a)29.68 (n/a)19.38 (n/a)15.82 (n/a)14.45 (n/a)6.47 (n/a)

test_gemm[M_192-K_384-N_64-num_aie_columns_4-b_col_maj_True-c_col_maj_True-m_48-k_96-n_16-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.62 (+40.30%)0.44 (+46.70%)0.42 (+19.01%)0.32 (+241.45%)0.12 (-9.78%)693.40 (-70.71%)532.58 (-46.59%)531.80 (-15.96%)357.70 (-28.72%)133.36 (-82.83%)26.38 (+40.30%)18.72 (+46.70%)17.75 (+19.01%)13.61 (+241.45%)5.08 (-9.78%)
3ad2eb2 — 2026-08-29 00:27:500.44 (n/a)0.30 (n/a)0.35 (n/a)0.09 (n/a)0.13 (n/a)2367.70 (n/a)997.10 (n/a)632.80 (n/a)501.80 (n/a)776.90 (n/a)18.81 (n/a)12.76 (n/a)14.91 (n/a)3.99 (n/a)5.63 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_1-b_col_maj_False-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.31 (-0.50%)0.31 (+0.81%)0.31 (+0.22%)0.30 (+3.91%)0.00 (-52.74%)83584.80 (-3.76%)82379.64 (-0.85%)82323.50 (-0.22%)80859.80 (+0.50%)1086.23 (-54.45%)212.46 (-0.50%)208.57 (+0.81%)208.69 (+0.22%)205.54 (+3.91%)2.76 (-52.74%)
3ad2eb2 — 2026-08-29 00:27:500.31 (n/a)0.30 (n/a)0.31 (n/a)0.29 (n/a)0.01 (n/a)86854.60 (n/a)83088.32 (n/a)82506.20 (n/a)80455.40 (n/a)2384.65 (n/a)213.53 (n/a)206.90 (n/a)208.23 (n/a)197.80 (n/a)5.84 (n/a)

test_gemm[M_2048-K_2048-N_2048-num_aie_columns_2-b_col_maj_True-c_col_maj_False-m_64-k_64-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:361.02 (+0.20%)1.00 (+1.56%)1.01 (-0.06%)0.98 (+4.79%)0.02 (-40.55%)25796.40 (-4.57%)25151.50 (-1.62%)24968.20 (+0.06%)24594.60 (-0.20%)605.92 (-43.20%)698.52 (+0.20%)683.37 (+1.56%)688.07 (-0.06%)665.98 (+4.79%)16.38 (-40.55%)
3ad2eb2 — 2026-08-29 00:27:501.02 (n/a)0.99 (n/a)1.01 (n/a)0.93 (n/a)0.04 (n/a)27032.40 (n/a)25565.84 (n/a)24953.90 (n/a)24643.50 (n/a)1066.75 (n/a)697.14 (n/a)672.90 (n/a)688.46 (n/a)635.53 (n/a)27.55 (n/a)

test_gemm[M_384-K_1536-N_1792-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_32-k_48-n_64-trace_size_0-partition_N_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:363.69 (+9.43%)2.63 (-1.70%)2.32 (-16.05%)1.58 (-16.46%)0.88 (+30.79%)5104.60 (+19.71%)3365.68 (+5.91%)3470.00 (+19.12%)2187.40 (-8.62%)1171.81 (+39.31%)966.42 (+9.43%)690.20 (-1.70%)609.20 (-16.05%)414.13 (-16.46%)229.98 (+30.79%)
3ad2eb2 — 2026-08-29 00:27:503.37 (n/a)2.68 (n/a)2.77 (n/a)1.89 (n/a)0.67 (n/a)4264.20 (n/a)3177.78 (n/a)2913.00 (n/a)2393.70 (n/a)841.13 (n/a)883.10 (n/a)702.16 (n/a)725.70 (n/a)495.74 (n/a)175.84 (n/a)

test_gemm[M_64-K_512-N_256-num_aie_columns_4-b_col_maj_True-c_col_maj_False-m_16-k_64-n_64-trace_size_0-partition_N_4]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.30 (+58.67%)0.21 (+18.73%)0.18 (-2.85%)0.17 (+14.49%)0.06 (+208.66%)7443.00 (-12.66%)6171.84 (-12.34%)6925.00 (+2.94%)4086.80 (-36.97%)1403.93 (+67.04%)16.42 (+58.67%)11.43 (+18.73%)9.69 (-2.85%)9.02 (+14.49%)3.10 (+208.66%)
3ad2eb2 — 2026-08-29 00:27:500.19 (n/a)0.18 (n/a)0.19 (n/a)0.15 (n/a)0.02 (n/a)8521.70 (n/a)7040.34 (n/a)6727.30 (n/a)6484.40 (n/a)840.47 (n/a)10.35 (n/a)9.63 (n/a)9.98 (n/a)7.88 (n/a)1.00 (n/a)
iron/operators/gemv

test_gemv[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.15 (n/a)0.12 (n/a)0.12 (n/a)0.07 (n/a)0.03 (n/a)0.15 (n/a)0.11 (n/a)0.12 (n/a)0.07 (n/a)0.03 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:363.97 (n/a)3.73 (n/a)3.71 (n/a)3.54 (n/a)0.16 (n/a)3.97 (n/a)3.73 (n/a)3.71 (n/a)3.54 (n/a)0.16 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_2-tile_size_input_1-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:367.34 (-1.00%)6.21 (-11.21%)5.68 (-20.60%)5.67 (-8.72%)0.77 (+56.78%)7.33 (-1.00%)6.20 (-11.21%)5.67 (-20.60%)5.67 (-8.72%)0.77 (+56.78%)
3ad2eb2 — 2026-08-29 00:27:507.41 (n/a)6.99 (n/a)7.15 (n/a)6.21 (n/a)0.49 (n/a)7.41 (n/a)6.99 (n/a)7.14 (n/a)6.21 (n/a)0.49 (n/a)

test_gemv[M_2048-K_8192-num_aie_columns_4-tile_size_input_1-tile_size_output_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:3613.91 (+5.79%)10.27 (+7.71%)9.53 (+8.98%)7.27 (-1.00%)2.52 (+13.72%)13.90 (+5.79%)10.27 (+7.71%)9.53 (+8.98%)7.27 (-1.00%)2.51 (+13.72%)
3ad2eb2 — 2026-08-29 00:27:5013.15 (n/a)9.54 (n/a)8.75 (n/a)7.34 (n/a)2.21 (n/a)13.14 (n/a)9.53 (n/a)8.74 (n/a)7.34 (n/a)2.21 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:363.77 (n/a)3.60 (n/a)3.63 (n/a)3.33 (n/a)0.17 (n/a)3.76 (n/a)3.60 (n/a)3.62 (n/a)3.33 (n/a)0.17 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_2-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:366.97 (-2.69%)5.83 (-2.73%)5.64 (-2.13%)4.80 (-10.70%)0.80 (+14.77%)6.96 (-2.69%)5.83 (-2.73%)5.63 (-2.13%)4.80 (-10.70%)0.79 (+14.77%)
3ad2eb2 — 2026-08-29 00:27:507.16 (n/a)5.99 (n/a)5.76 (n/a)5.37 (n/a)0.69 (n/a)7.15 (n/a)5.99 (n/a)5.75 (n/a)5.37 (n/a)0.69 (n/a)

test_gemv[M_8192-K_2048-num_aie_columns_4-tile_size_input_4-tile_size_output_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:369.92 (+1.17%)8.40 (-0.69%)8.27 (+1.72%)7.36 (-4.78%)0.93 (+15.70%)9.91 (+1.17%)8.39 (-0.69%)8.26 (+1.72%)7.36 (-4.78%)0.93 (+15.70%)
3ad2eb2 — 2026-08-29 00:27:509.80 (n/a)8.46 (n/a)8.13 (n/a)7.73 (n/a)0.80 (n/a)9.80 (n/a)8.45 (n/a)8.12 (n/a)7.73 (n/a)0.80 (n/a)

test_gemv_batched[M_1024-K_1024-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:362.76 (-9.66%)2.11 (+11.24%)2.66 (+33.84%)1.19 (+1.82%)0.83 (+6.71%)2.75 (-9.66%)2.11 (+11.24%)2.66 (+33.84%)1.19 (+1.82%)0.83 (+6.71%)
3ad2eb2 — 2026-08-29 00:27:503.05 (n/a)1.90 (n/a)1.99 (n/a)1.17 (n/a)0.78 (n/a)3.05 (n/a)1.90 (n/a)1.98 (n/a)1.17 (n/a)0.78 (n/a)

test_gemv_batched[M_1026-K_64-num_aie_columns_1-tile_size_input_1-tile_size_output_2-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.52 (-19.73%)0.38 (+0.68%)0.40 (+13.35%)0.14 (+88.53%)0.15 (-28.65%)0.51 (-19.73%)0.38 (+0.68%)0.39 (+13.35%)0.14 (+88.53%)0.15 (-28.65%)
3ad2eb2 — 2026-08-29 00:27:500.64 (n/a)0.38 (n/a)0.35 (n/a)0.07 (n/a)0.21 (n/a)0.63 (n/a)0.37 (n/a)0.34 (n/a)0.07 (n/a)0.21 (n/a)

test_gemv_batched[M_256-K_128-num_aie_columns_1-tile_size_input_1-tile_size_output_256-num_batches_4]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:360.59 (+11.55%)0.30 (-23.36%)0.29 (-39.57%)0.08 (-7.43%)0.21 (+16.44%)0.58 (+11.55%)0.30 (-23.36%)0.28 (-39.57%)0.07 (-7.43%)0.21 (+16.44%)
3ad2eb2 — 2026-08-29 00:27:500.52 (n/a)0.40 (n/a)0.47 (n/a)0.08 (n/a)0.18 (n/a)0.52 (n/a)0.39 (n/a)0.47 (n/a)0.08 (n/a)0.18 (n/a)

test_gemv_batched[M_64-K_1536-num_aie_columns_1-tile_size_input_1-tile_size_output_64-num_batches_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Throughput (max)Throughput (mean)Throughput (median)Throughput (min)Throughput (stddev)
6ea8fb8 — 2026-08-29 03:22:362.38 (-11.30%)1.20 (-25.63%)0.72 (-62.43%)0.42 (-6.56%)0.86 (-22.43%)2.34 (-11.30%)1.18 (-25.63%)0.70 (-62.43%)0.41 (-6.56%)0.84 (-22.43%)
3ad2eb2 — 2026-08-29 00:27:502.68 (n/a)1.62 (n/a)1.91 (n/a)0.45 (n/a)1.11 (n/a)2.64 (n/a)1.59 (n/a)1.88 (n/a)0.44 (n/a)1.09 (n/a)

test_gemv_gelu[M_128-K_128-num_aie_columns_1-tile_size_input_32-tile_size_output_128]

No metrics available.

test_gemv_gelu[M_2048-K_8192-num_aie_columns_1-tile_size_input_1-tile_size_output_2048]

No metrics available.

test_gemv_gelu[M_8192-K_2048-num_aie_columns_1-tile_size_input_4-tile_size_output_1024]

No metrics available.

iron/operators/layer_norm

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)508.50 (n/a)380.64 (n/a)386.50 (n/a)223.50 (n/a)129.38 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)464.90 (n/a)340.98 (n/a)306.70 (n/a)247.90 (n/a)99.82 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)613.20 (n/a)366.86 (n/a)265.00 (n/a)225.50 (n/a)167.84 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)520.00 (n/a)353.30 (n/a)343.00 (n/a)267.90 (n/a)101.98 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.02 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)670.50 (n/a)430.40 (n/a)317.10 (n/a)285.20 (n/a)185.50 (n/a)

test_layer_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)635.50 (n/a)453.20 (n/a)461.50 (n/a)283.00 (n/a)126.34 (n/a)
iron/operators/leaky_relu

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (+14.06%)0.02 (+1.44%)0.03 (+19.99%)0.00 (-71.70%)0.01 (+88.00%)1840.30 (+253.29%)607.02 (+67.09%)256.10 (-16.66%)230.90 (-12.34%)696.50 (+489.52%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)520.90 (n/a)363.30 (n/a)307.30 (n/a)263.40 (n/a)118.15 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (+19.67%)0.03 (+17.08%)0.03 (+33.70%)0.01 (-8.27%)0.01 (+61.45%)559.40 (+9.02%)334.60 (-8.09%)234.50 (-25.20%)229.30 (-16.41%)149.12 (+40.64%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)513.10 (n/a)364.06 (n/a)313.50 (n/a)274.30 (n/a)106.03 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-alpha_0.25]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-12.84%)0.02 (+1.63%)0.02 (+25.46%)0.02 (+4.14%)0.01 (-18.29%)522.80 (-3.97%)388.66 (-3.58%)354.00 (-20.29%)262.60 (+14.72%)121.54 (-5.06%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)544.40 (n/a)403.10 (n/a)444.10 (n/a)228.90 (n/a)128.02 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-2.92%)0.03 (+26.23%)0.03 (+59.33%)0.02 (+54.74%)0.00 (-50.12%)395.60 (-35.37%)306.28 (-28.18%)293.40 (-37.23%)249.70 (+3.01%)53.95 (-64.98%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)612.10 (n/a)426.44 (n/a)467.40 (n/a)242.40 (n/a)154.05 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (+30.30%)0.02 (+23.23%)0.02 (+16.87%)0.01 (-11.91%)0.01 (+98.44%)652.40 (+13.54%)435.12 (-8.63%)433.40 (-14.43%)231.20 (-23.24%)188.24 (+82.81%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)574.60 (n/a)476.20 (n/a)506.50 (n/a)301.20 (n/a)102.97 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (+17.46%)0.02 (+22.15%)0.02 (+9.28%)0.01 (+84.12%)0.00 (-38.15%)547.80 (-45.69%)475.94 (-23.03%)482.20 (-8.50%)403.90 (-14.86%)60.39 (-72.75%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.01 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)1008.60 (n/a)618.36 (n/a)527.00 (n/a)474.40 (n/a)221.65 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-8.87%)0.02 (+10.10%)0.03 (+41.68%)0.01 (+14.56%)0.01 (-12.55%)557.60 (-12.71%)393.66 (-11.89%)302.50 (-29.44%)292.20 (+9.73%)133.38 (-18.76%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)638.80 (n/a)446.80 (n/a)428.70 (n/a)266.30 (n/a)164.18 (n/a)

test_leaky_relu[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-alpha_0.01]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-9.63%)0.02 (+16.30%)0.02 (+45.44%)0.01 (-51.87%)0.01 (+27.89%)1350.00 (+107.76%)529.32 (+9.68%)330.70 (-31.25%)299.60 (+10.68%)459.18 (+232.34%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)649.80 (n/a)482.60 (n/a)481.00 (n/a)270.70 (n/a)138.17 (n/a)
iron/operators/mem_copy

test_mem_copy[input_length_2048-num_cores_1-num_channels_1-bypass_False-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-14.05%)0.02 (-17.27%)0.02 (-34.15%)0.02 (+5.04%)0.01 (-13.46%)510.50 (-4.79%)417.18 (+19.46%)480.10 (+51.88%)301.00 (+16.35%)105.04 (-6.88%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)536.20 (n/a)349.22 (n/a)316.10 (n/a)258.70 (n/a)112.80 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_1-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (-19.19%)0.02 (-30.50%)0.02 (-41.30%)0.01 (-8.72%)0.01 (-16.43%)607.20 (+9.54%)450.52 (+41.76%)453.70 (+70.37%)226.80 (+23.73%)143.99 (+1.85%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)554.30 (n/a)317.80 (n/a)266.30 (n/a)183.30 (n/a)141.37 (n/a)

test_mem_copy[input_length_2048-num_cores_2-num_channels_2-bypass_False-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-32.21%)0.02 (-15.86%)0.03 (-23.40%)0.02 (+25.25%)0.00 (-67.91%)431.60 (-20.16%)342.98 (+4.82%)320.90 (+30.55%)287.00 (+47.48%)55.28 (-62.25%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)540.60 (n/a)327.20 (n/a)245.80 (n/a)194.60 (n/a)146.43 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_1-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (+21.85%)0.02 (+13.25%)0.02 (-5.22%)0.01 (-20.53%)0.01 (+66.84%)790.80 (+25.82%)472.18 (-2.68%)486.20 (+5.51%)259.90 (-17.93%)215.65 (+53.81%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)628.50 (n/a)485.16 (n/a)460.80 (n/a)316.70 (n/a)140.21 (n/a)

test_mem_copy[input_length_2048-num_cores_4-num_channels_2-bypass_False-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (+10.26%)0.02 (+11.94%)0.02 (+7.10%)0.01 (+0.92%)0.00 (+39.78%)582.90 (-0.92%)457.78 (-9.19%)484.20 (-6.63%)347.20 (-9.30%)100.32 (+22.05%)
3ad2eb2 — 2026-08-29 00:27:500.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.00 (n/a)588.30 (n/a)504.12 (n/a)518.60 (n/a)382.80 (n/a)82.20 (n/a)

test_mem_copy[input_length_2048-num_cores_8-num_channels_2-bypass_False-tile_size_256]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (+6.85%)0.02 (-3.91%)0.02 (-1.30%)0.01 (-18.27%)0.01 (+28.39%)661.60 (+22.36%)484.62 (+8.33%)509.50 (+1.31%)279.00 (-6.41%)150.96 (+42.87%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)540.70 (n/a)447.36 (n/a)502.90 (n/a)298.10 (n/a)105.67 (n/a)
iron/operators/rms_norm

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-9.99%)0.02 (-15.70%)0.02 (-27.19%)0.01 (-18.88%)0.01 (-9.27%)709.50 (+23.26%)414.44 (+20.17%)356.70 (+37.35%)249.90 (+11.12%)190.13 (+24.24%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)575.60 (n/a)344.88 (n/a)259.70 (n/a)224.90 (n/a)153.04 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_1-tile_size_2048-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (+0.65%)0.04 (+11.39%)0.05 (+24.61%)0.02 (-1.04%)0.01 (+17.35%)534.60 (+1.06%)313.14 (-8.19%)247.90 (-19.75%)246.90 (-0.64%)125.11 (+13.47%)
3ad2eb2 — 2026-08-29 00:27:500.05 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.01 (n/a)529.00 (n/a)341.06 (n/a)308.90 (n/a)248.50 (n/a)110.26 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (+3.42%)0.02 (-31.83%)0.02 (-45.22%)0.01 (-26.49%)0.01 (+43.48%)628.60 (+36.03%)480.56 (+56.18%)516.50 (+82.57%)239.10 (-3.28%)144.58 (+64.98%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)462.10 (n/a)307.70 (n/a)282.90 (n/a)247.20 (n/a)87.64 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_1-num_channels_2-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (+37.57%)0.04 (+32.74%)0.04 (+48.70%)0.02 (+26.06%)0.01 (+40.03%)444.40 (-20.67%)307.56 (-23.27%)242.60 (-32.76%)194.60 (-27.31%)120.60 (-14.66%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)560.20 (n/a)400.82 (n/a)360.80 (n/a)267.70 (n/a)141.31 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (+4.62%)0.02 (-18.14%)0.02 (-31.80%)0.01 (-29.89%)0.01 (+85.33%)560.90 (+42.65%)387.22 (+33.00%)409.50 (+46.62%)235.50 (-4.42%)138.60 (+135.52%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.00 (n/a)393.20 (n/a)291.14 (n/a)279.30 (n/a)246.40 (n/a)58.85 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_1-tile_size_1024-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.05 (+2.72%)0.03 (-16.96%)0.03 (-23.69%)0.02 (-10.33%)0.01 (+23.31%)608.80 (+11.52%)381.34 (+27.89%)316.40 (+31.01%)207.30 (-2.68%)179.71 (+28.83%)
3ad2eb2 — 2026-08-29 00:27:500.05 (n/a)0.04 (n/a)0.04 (n/a)0.02 (n/a)0.01 (n/a)545.90 (n/a)298.18 (n/a)241.50 (n/a)213.00 (n/a)139.49 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-5.60%)0.02 (-6.54%)0.03 (-5.72%)0.01 (+26.32%)0.01 (-9.66%)552.50 (-20.83%)362.04 (+1.37%)304.40 (+6.06%)247.10 (+5.92%)131.71 (-31.42%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.03 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)697.90 (n/a)357.14 (n/a)287.00 (n/a)233.30 (n/a)192.05 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_2-num_channels_2-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (+34.15%)0.02 (-5.55%)0.02 (-37.75%)0.01 (-36.60%)0.01 (+53.11%)1283.40 (+57.74%)576.46 (+31.22%)481.60 (+60.64%)217.60 (-25.45%)426.93 (+88.13%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.03 (n/a)0.01 (n/a)0.01 (n/a)813.60 (n/a)439.30 (n/a)299.80 (n/a)291.90 (n/a)226.93 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.04 (-5.09%)0.02 (-0.67%)0.02 (+7.92%)0.01 (-2.53%)0.01 (-9.14%)643.50 (+2.60%)467.30 (-1.16%)489.30 (-7.33%)219.00 (+5.34%)158.54 (-2.93%)
3ad2eb2 — 2026-08-29 00:27:500.04 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)627.20 (n/a)472.80 (n/a)528.00 (n/a)207.90 (n/a)163.34 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_1-tile_size_512-weighted_True]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-3.64%)0.02 (-13.32%)0.02 (-20.39%)0.01 (-19.77%)0.01 (-1.41%)660.60 (+24.64%)437.06 (+17.22%)407.90 (+25.62%)281.40 (+3.76%)146.29 (+29.47%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.03 (n/a)0.03 (n/a)0.02 (n/a)0.01 (n/a)530.00 (n/a)372.84 (n/a)324.70 (n/a)271.20 (n/a)112.99 (n/a)

test_rms_norm[input_length_2048-num_aie_columns_4-num_channels_2-tile_size_256-weighted_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.03 (-4.10%)0.02 (-3.94%)0.02 (+14.23%)0.01 (-7.56%)0.01 (-9.33%)739.20 (+8.18%)479.36 (+3.88%)418.70 (-12.46%)296.70 (+4.25%)184.25 (+8.70%)
3ad2eb2 — 2026-08-29 00:27:500.03 (n/a)0.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)683.30 (n/a)461.46 (n/a)478.30 (n/a)284.60 (n/a)169.50 (n/a)
iron/operators/rope

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.38 (+11.20%)0.25 (+9.61%)0.21 (+6.51%)0.11 (-27.83%)0.11 (+42.09%)860.30 (+38.58%)473.40 (+0.57%)472.20 (-6.10%)260.80 (-10.07%)237.90 (+74.70%)
3ad2eb2 — 2026-08-29 00:27:500.34 (n/a)0.23 (n/a)0.20 (n/a)0.16 (n/a)0.07 (n/a)620.80 (n/a)470.72 (n/a)502.90 (n/a)290.00 (n/a)136.18 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.45 (+21.04%)0.29 (+9.27%)0.21 (-29.79%)0.17 (+14.46%)0.13 (+43.31%)572.50 (-12.64%)394.24 (-4.66%)466.60 (+42.43%)219.20 (-17.41%)157.66 (-3.67%)
3ad2eb2 — 2026-08-29 00:27:500.37 (n/a)0.27 (n/a)0.30 (n/a)0.15 (n/a)0.09 (n/a)655.30 (n/a)413.52 (n/a)327.60 (n/a)265.40 (n/a)163.67 (n/a)

test_rope[rows_32-cols_512-angle_rows_32-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.43 (+50.77%)0.30 (+46.66%)0.32 (+64.06%)0.17 (+0.63%)0.12 (+141.76%)581.80 (-0.63%)369.94 (-24.69%)305.80 (-39.05%)226.70 (-33.67%)157.02 (+60.49%)
3ad2eb2 — 2026-08-29 00:27:500.29 (n/a)0.21 (n/a)0.20 (n/a)0.17 (n/a)0.05 (n/a)585.50 (n/a)491.24 (n/a)501.70 (n/a)341.80 (n/a)97.84 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_1-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.31 (+21.69%)0.26 (+48.61%)0.28 (+82.52%)0.16 (+12.64%)0.06 (+27.35%)468.50 (-11.22%)303.78 (-31.96%)266.50 (-45.21%)234.90 (-17.81%)95.86 (-3.71%)
3ad2eb2 — 2026-08-29 00:27:500.26 (n/a)0.17 (n/a)0.15 (n/a)0.14 (n/a)0.05 (n/a)527.70 (n/a)446.48 (n/a)486.40 (n/a)285.80 (n/a)99.55 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_2-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.31 (+7.48%)0.21 (-13.35%)0.17 (-33.04%)0.12 (-17.29%)0.09 (+50.91%)610.40 (+20.90%)401.60 (+24.66%)443.60 (+49.36%)234.60 (-6.94%)160.31 (+53.11%)
3ad2eb2 — 2026-08-29 00:27:500.29 (n/a)0.24 (n/a)0.25 (n/a)0.15 (n/a)0.06 (n/a)504.90 (n/a)322.16 (n/a)297.00 (n/a)252.10 (n/a)104.70 (n/a)

test_rope[rows_32-cols_512-angle_rows_8-aie_columns_4-method_type_0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.43 (+43.26%)0.24 (+32.84%)0.17 (+12.82%)0.13 (+41.06%)0.13 (+50.14%)548.60 (-29.11%)374.90 (-22.61%)439.80 (-11.37%)169.70 (-30.19%)166.70 (-23.43%)
3ad2eb2 — 2026-08-29 00:27:500.30 (n/a)0.18 (n/a)0.15 (n/a)0.10 (n/a)0.09 (n/a)773.90 (n/a)484.46 (n/a)496.20 (n/a)243.10 (n/a)217.71 (n/a)
iron/operators/softmax

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_1024]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.48 (+29.94%)0.34 (+30.81%)0.29 (+17.78%)0.22 (+8.92%)0.12 (+83.73%)588.90 (-8.19%)422.22 (-19.30%)452.60 (-15.10%)270.90 (-23.06%)138.59 (+30.76%)
3ad2eb2 — 2026-08-29 00:27:500.37 (n/a)0.26 (n/a)0.25 (n/a)0.20 (n/a)0.06 (n/a)641.40 (n/a)523.18 (n/a)533.10 (n/a)352.10 (n/a)105.98 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.57 (+33.76%)0.31 (+20.27%)0.26 (+17.26%)0.21 (+23.90%)0.15 (+32.51%)638.00 (-19.29%)475.54 (-17.60%)503.70 (-14.71%)230.10 (-25.24%)149.30 (-30.11%)
3ad2eb2 — 2026-08-29 00:27:500.43 (n/a)0.26 (n/a)0.22 (n/a)0.17 (n/a)0.11 (n/a)790.50 (n/a)577.08 (n/a)590.60 (n/a)307.80 (n/a)213.62 (n/a)

test_softmax[input_length_32768-num_aie_columns_2-num_channels_2-tile_size_512]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.47 (+11.14%)0.35 (+21.49%)0.38 (-3.82%)0.17 (+135.22%)0.12 (-23.04%)782.10 (-57.49%)429.60 (-40.82%)344.90 (+3.98%)276.60 (-10.02%)208.53 (-68.34%)
3ad2eb2 — 2026-08-29 00:27:500.43 (n/a)0.29 (n/a)0.40 (n/a)0.07 (n/a)0.16 (n/a)1839.60 (n/a)725.98 (n/a)331.70 (n/a)307.40 (n/a)658.64 (n/a)
iron/operators/strided_copy

test_strided_copy[chunked_transfer]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.01 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)587.70 (n/a)476.50 (n/a)539.60 (n/a)289.90 (n/a)133.89 (n/a)

test_strided_copy[contiguous]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)0.01 (n/a)2045.00 (n/a)694.26 (n/a)419.80 (n/a)217.90 (n/a)764.37 (n/a)

test_strided_copy[four_channels]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (n/a)0.02 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)294.10 (n/a)264.26 (n/a)277.40 (n/a)208.20 (n/a)35.85 (n/a)

test_strided_copy[kv_slot0]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.58 (n/a)0.51 (n/a)0.48 (n/a)0.46 (n/a)0.06 (n/a)288.70 (n/a)262.80 (n/a)277.20 (n/a)229.30 (n/a)30.58 (n/a)

test_strided_copy[kv_slot5]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.66 (n/a)0.46 (n/a)0.45 (n/a)0.23 (n/a)0.18 (n/a)569.60 (n/a)329.76 (n/a)295.10 (n/a)199.60 (n/a)151.00 (n/a)

test_strided_copy[kv_slot5_four_channels]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.44 (n/a)0.37 (n/a)0.44 (n/a)0.23 (n/a)0.10 (n/a)572.40 (n/a)380.42 (n/a)302.30 (n/a)297.40 (n/a)121.44 (n/a)

test_strided_copy[kv_slot5_two_channels]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.51 (n/a)0.43 (n/a)0.46 (n/a)0.29 (n/a)0.09 (n/a)458.80 (n/a)318.58 (n/a)290.10 (n/a)257.60 (n/a)80.22 (n/a)

test_strided_copy[kv_slot_last]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.57 (n/a)0.43 (n/a)0.46 (n/a)0.27 (n/a)0.11 (n/a)486.80 (n/a)325.40 (n/a)289.10 (n/a)230.70 (n/a)98.13 (n/a)

test_strided_copy[two_channels]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)413.00 (n/a)300.38 (n/a)290.20 (n/a)237.50 (n/a)67.12 (n/a)

test_strided_copy[two_channels_chunked]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.02 (n/a)0.01 (n/a)0.01 (n/a)0.01 (n/a)0.00 (n/a)518.20 (n/a)356.90 (n/a)294.30 (n/a)264.30 (n/a)112.91 (n/a)

test_transfer_size_not_dividing_per_channel_share_is_rejected[iter0]

No metrics available.

test_transfer_size_not_dividing_per_channel_share_is_rejected[iter1]

No metrics available.

test_transfer_size_not_dividing_per_channel_share_is_rejected[iter2]

No metrics available.

test_transfer_size_not_dividing_per_channel_share_is_rejected[iter3]

No metrics available.

test_transfer_size_not_dividing_per_channel_share_is_rejected[iter4]

No metrics available.

iron/operators/swiglu_decode

test_swiglu_decode[embedding_dim_1024-hidden_dim_3584]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.00 (+16.67%)0.00 (+36.84%)0.00 (+66.67%)0.00 (+50.00%)0.00 (-27.63%)16112.95 (-28.86%)8839.95 (-36.33%)7639.30 (-50.90%)5587.46 (-14.13%)4162.51 (-39.24%)
3ad2eb2 — 2026-08-29 00:27:500.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)22648.63 (n/a)13884.09 (n/a)15557.60 (n/a)6506.70 (n/a)6850.41 (n/a)

test_swiglu_decode[embedding_dim_2048-hidden_dim_2048]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.00 (+0.00%)0.00 (-23.81%)0.00 (-54.55%)0.00 (+0.00%)0.00 (-14.76%)22413.57 (+0.26%)15665.92 (+25.77%)17940.94 (+135.76%)7236.92 (-2.52%)6541.59 (-6.71%)
3ad2eb2 — 2026-08-29 00:27:500.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)0.00 (n/a)22356.06 (n/a)12456.27 (n/a)7609.99 (n/a)7424.25 (n/a)7012.19 (n/a)
iron/operators/swiglu_prefill

test_swiglu_prefill[seq_len_256-embedding_dim_2048-hidden_dim_2048-prio_accuracy_False]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:360.14 (-10.12%)0.09 (-26.41%)0.08 (-38.03%)0.07 (+3.36%)0.03 (-17.52%)29681.07 (-3.21%)24198.95 (+32.27%)25622.37 (+61.33%)15236.63 (+11.26%)5638.21 (-19.26%)
3ad2eb2 — 2026-08-29 00:27:500.15 (n/a)0.12 (n/a)0.13 (n/a)0.07 (n/a)0.03 (n/a)30665.29 (n/a)18295.30 (n/a)15881.58 (n/a)13695.18 (n/a)6982.78 (n/a)
iron/operators/transpose

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:361.74 (+2.15%)1.30 (+18.82%)1.49 (+48.00%)0.51 (-29.73%)0.50 (+34.53%)1030.40 (+42.30%)494.38 (-4.78%)351.40 (-32.44%)300.90 (-2.11%)306.21 (+103.32%)
3ad2eb2 — 2026-08-29 00:27:501.71 (n/a)1.09 (n/a)1.01 (n/a)0.72 (n/a)0.37 (n/a)724.10 (n/a)519.18 (n/a)520.10 (n/a)307.40 (n/a)150.60 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8-num_batches_2]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:362.58 (+8.44%)1.75 (+29.17%)1.53 (+12.20%)1.46 (+386.57%)0.47 (-40.00%)720.20 (-79.45%)625.86 (-50.85%)683.40 (-10.88%)407.00 (-7.77%)126.32 (-90.03%)
3ad2eb2 — 2026-08-29 00:27:502.38 (n/a)1.36 (n/a)1.37 (n/a)0.30 (n/a)0.78 (n/a)3504.30 (n/a)1273.28 (n/a)766.80 (n/a)441.30 (n/a)1267.21 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_1-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-25 20:01:371.58 (-2.67%)1.21 (+5.65%)1.29 (+41.89%)0.66 (-20.05%)0.38 (-2.53%)792.90 (+25.08%)479.12 (-3.80%)406.80 (-29.52%)331.10 (+2.76%)189.18 (+26.62%)
4bb8427 — 2026-06-23 22:46:491.63 (n/a)1.14 (n/a)0.91 (n/a)0.83 (n/a)0.38 (n/a)633.90 (n/a)498.02 (n/a)577.20 (n/a)322.20 (n/a)149.40 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8-num_batches_1]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
6ea8fb8 — 2026-08-29 03:22:361.66 (-16.36%)0.92 (-25.46%)0.92 (-8.98%)0.26 (-65.47%)0.50 (+2.01%)1992.60 (+189.62%)816.64 (+72.08%)571.70 (+9.86%)315.70 (+19.58%)670.66 (+302.36%)
3ad2eb2 — 2026-08-29 00:27:501.99 (n/a)1.24 (n/a)1.01 (n/a)0.76 (n/a)0.49 (n/a)688.00 (n/a)474.58 (n/a)520.40 (n/a)264.00 (n/a)166.68 (n/a)

test_transpose[M_2048-N_64-aie_columns_1-channels_2-m_64-n_64-s_8]

Commit/Date Bandwidth (max)Bandwidth (mean)Bandwidth (median)Bandwidth (min)Bandwidth (stddev)Latency (max)Latency (mean)Latency (median)Latency (min)Latency (stddev)
4bb8427 — 2026-06-25 20:01:371.75 (-15.44%)1.32 (-0.50%)1.36 (+22.47%)0.77 (-19.20%)0.36 (-21.75%)678.10 (+23.76%)430.02 (-0.09%)385.10 (-18.34%)300.10 (+18.29%)146.96 (+19.46%)
4bb8427 — 2026-06-23 22:46:492.07 (n/a)1.32 (n/a)1.11 (n/a)0.96 (n/a)0.46 (n/a)547.90 (n/a)430.40 (n/a)471.60 (n/a)253.70 (n/a)123.02 (n/a)
Phoenix - Examples

IRON

Tested on 2026_08_29_03_29_03 at commit 6ea8fb8.

Trends:

IRON Trends

@andrej
andrej added this pull request to the merge queue Aug 29, 2026
Merged via the queue into amd:devel with commit a15c8cd Aug 29, 2026
5 checks passed
@atassis
atassis deleted the pr/strided-copy-transfer-size branch August 29, 2026 21:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants