# The WAI_OPS_GPU_REPORT lines of the runs on which determinism-tiers §5 claims
# decode-equivalence for the wai-ops/1-i32 graphs of wai/wai-ops-gpu, on this revision's
# kernel text (wai/wai-ops-gpu/src/wgsl/): runs 16 to 18. Each line is as the tests wrote it,
# except that its `adapter="..."` field, the name the backend reports for the device, is removed;
# the backend, PCI vendor and device, driver and driver version are as reported.
#
# Runs 1 and 2 are kept as the record of the kernel text before the narrow kernels (conv_n,
# conv_t_n of ops64.wgsl), and runs 3 to 5 as that of the text with the narrow kernels and
# before the run-time and four-channel forms (mag, conv_r, conv_t_r, conv_r4, conv_t_r4, igdn_r),
# runs 6 to 9 as that of the text with those forms and before the tiled kernels (tile.wgsl:
# conv_tile, igdn_tile), runs 10 to 12 as that of the text with the tiled kernels and before
# the read forms (out.wgsl: rgb8, lows), and runs 13 to 15 as that of the text with the read forms
# and before the tiled convolution's bound per tile and stage (tile.wgsl: mag_tile, tile_chunk),
# which this revision added; they claim nothing for this text.
#
# Run 1. The forge's run 10236, job 46928, of .github/workflows/wai-ops-gpu.yml, on the
#   `macos` runner (arm64 desktop, macOS 27.0.1, Metal; the backend reports no PCI identity,
#   vendor and device 0x0), 2026-10-22, on the kernel text of commit c47a58cca, which the
#   kernels before the narrow ones were byte for byte. rustc 1.98.0, wgpu and naga 30.0.1,
#   release.
# Run 2. `WAI_OPS_GPU_REQUIRE=1 cargo test --locked --release -- --nocapture
#   --test-threads=1` in wai/wai-ops-gpu, a Linux x86_64 host (kernel 7.0), 2026-10-24, at
#   commit 2aa4e54d3 of the change that added this file. rustc 1.98.0, wgpu and naga
#   30.0.1, release: every test of the crate ran and passed. The capability corpora (wai-rs
#   tests/ops_gpu.rs) were not run on this host.
# Runs 3 to 5 are of the kernel text before this revision's, by hand on 2026-11-11 through rb, rustc
#   1.99.1, wgpu and naga 30.0.1, release builds, each the same four commands:
#   `WAI_OPS_GPU_REQUIRE=1 cargo test --locked --release -- --nocapture --test-threads=1` and
#   `cargo clippy --locked --release --all-targets -- -D warnings` in wai/wai-ops-gpu (every
#   test passed, clippy clean); in wai/wai-rs, under WAI_OPS_GPU_REQUIRE=1, `cargo test --locked
#   --release --no-default-features --features ops_gpu,video_int,audio_int,splat_int,world_int
#   --test ops_gpu` (the jpegai/v1 inputs were not on the hosts: its line says it did not run)
#   and `cargo test --locked --release --no-default-features --features sink_decode --test
#   sink_decode`, whose `sink decode:` lines are the native sink's decoder (native-sink §13).
# Run 3. An arm64 laptop's integrated GPU, Metal, macOS 27.2.1 (the backend reports no PCI
#   identity).
# Run 4. An arm64 desktop's integrated GPU of an earlier generation, Metal, macOS 27.2.1.
# Run 5. A Linux x86_64 host (kernel 7.0.0), Vulkan: the integrated GPU of PCI vendor 0x1002
#   and llvmpipe. The native sink's decoder takes the first adapter that is not a software
#   one, the 0x1002 GPU.
#
# Runs 6 to 8 are of the kernel text before the tiled kernels, by hand on 2026-11-12 through rb,
#   rustc 1.99.1, wgpu and naga 30.0.1, release builds, the four commands of runs 3 to 5 with
#   the native sink's frame forms beside its decoder (`--test sink_decode --test sink_decode_frames`), and
#   `cargo clippy --locked --release --all-targets --no-default-features --features sink_decode
#   -- -D warnings` in wai/wai-rs: every test passed and both clippy runs were clean on each.
#   On run 8's host, under a load of some 60 on 32 cores, the measured-choice test failed once:
#   a 9 us graph's runs passed twice its median three times in a row, so its shape was measured
#   again on a decode the test required to reuse it. The rule now takes the median as at least
#   0.1 ms and the test allows a re-measure on that trigger; run 8's `sink decode` lines are of
#   that run of the native sink's tests, which passed, and runs 6 and 7 ran before the change.
# Run 6. An arm64 laptop's integrated GPU, Metal, macOS 27.2.1 (the laptop of run 3).
# Run 7. An arm64 desktop's integrated GPU of an earlier generation, Metal, macOS 27.2.1 (the
#   desktop of run 4).
# Run 8. A Linux x86_64 host, Vulkan: the integrated GPU of PCI vendor 0x1002 and llvmpipe (the
#   host of run 5).
#
# Run 1
ops-gpu arith mutant: key=gpu-metal-0 refused: the probe's word 3 differs from the host's on ProbeIn { a: 340282366920938463463374607431768211455, b: 34028236692
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=4207
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x0000
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases
# Run 2
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.2
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.7
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
# Run 3
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.4
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 8 of 8
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=178 edge=18 decodes=916 narrow_ops=155 finite_chunks=73 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
sink decode: planted "narrow chunk folded after 2K taps": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 4
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.7
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 8 of 8
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=178 edge=18 decodes=916 narrow_ops=155 finite_chunks=73 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
sink decode: planted "narrow chunk folded after 2K taps": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 5
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.5
ops-gpu corpus narrow: key=gpu-vulkan-1002 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.7
ops-gpu corpus narrow: key=swgpu-vulkan-10005 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu narrow mutants: key=gpu-vulkan-1002 killed 8 of 8
ops-gpu narrow mutants: key=swgpu-vulkan-10005 killed 8 of 8
ops-gpu narrow: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=178 edge=18 decodes=916 narrow_ops=155 finite_chunks=73 narrow_mismatches=0 wide_mismatches=0
ops-gpu narrow: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=178 edge=18 decodes=916 narrow_ops=155 finite_chunks=73 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
sink decode: planted "narrow chunk folded after 2K taps": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 6
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.0
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2580,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2580 us fills the reader's 1000000 us window only in 388 runs, more than 64"},"cpu":{"median_us":585,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 585 us fills the reader's 1000000 us window only in 1710 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"gpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":1619,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1619 us fills the reader's 1000000 us window only in 618 runs, more than 64"},"cpu":{"median_us":2259,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2259 us fills the reader's 1000000 us window only in 443 runs, more than 64"},"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=17 decodes_on_gpu=3 decodes_on_cpu=12 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=233 cpu_median_us=17
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=271 cpu_median_us=37
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=366 cpu_median_us=30
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=342 cpu_median_us=172
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=231 cpu_median_us=165
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=225 cpu_median_us=5
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=1000 cpu_median_us=1272
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=196 cpu_median_us=23
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=437 cpu_median_us=908
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=399 cpu_median_us=305
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=315 cpu_median_us=75
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=752 cpu_median_us=577
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=249 cpu_median_us=10
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=1636 cpu_median_us=2390
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=312 cpu_median_us=188
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=183 cpu_median_us=10
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=296 cpu_median_us=46
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 7
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.7
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":1887,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1887 us fills the reader's 1000000 us window only in 530 runs, more than 64"},"cpu":{"median_us":1861,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1861 us fills the reader's 1000000 us window only in 538 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"gpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2819,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2819 us fills the reader's 1000000 us window only in 355 runs, more than 64"},"cpu":{"median_us":5768,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 5768 us fills the reader's 1000000 us window only in 174 runs, more than 64"},"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=17 decodes_on_gpu=3 decodes_on_cpu=12 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=788 cpu_median_us=96
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=707 cpu_median_us=127
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1054 cpu_median_us=153
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1161 cpu_median_us=636
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=691 cpu_median_us=384
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=610 cpu_median_us=28
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=2285 cpu_median_us=4246
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=653 cpu_median_us=101
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=1328 cpu_median_us=2938
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1286 cpu_median_us=796
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=968 cpu_median_us=154
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=2458 cpu_median_us=1981
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=672 cpu_median_us=30
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=2313 cpu_median_us=6964
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1062 cpu_median_us=413
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=669 cpu_median_us=52
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=824 cpu_median_us=160
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 8
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=14 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=6 graphs_cpu=0 [] graphs_compared=6 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=1 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=10 graphs_cpu=0 [] graphs_compared=10 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-vulkan-1002 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus narrow: key=swgpu-vulkan-10005 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.3
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.0
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu narrow mutants: key=gpu-vulkan-1002 killed 19 of 19
ops-gpu narrow mutants: key=swgpu-vulkan-10005 killed 19 of 19
ops-gpu narrow: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=192 edge=18 runtime=14 decodes=962 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu narrow: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=192 edge=18 runtime=14 decodes=962 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
sink decode choice energy: reader="RAPL package energy (the whole package: every process on it)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":4507,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 4507 us fills the reader's 500000 us window only in 111 runs, more than 64"},"cpu":{"median_us":1178,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1178 us fills the reader's 500000 us window only in 425 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"gpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":6687,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 6687 us fills the reader's 500000 us window only in 75 runs, more than 64"},"cpu":{"median_us":12479,"runs":5,"energy_uj":807186,"energy_runs":41,"no_energy":null},"measured_now":true,"graphs":1}]}
sink decode choice: backend=vulkan shapes=17 decodes_on_gpu=2 decodes_on_cpu=13 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=511 cpu_median_us=32
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=697 cpu_median_us=65
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1526 cpu_median_us=71
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1124 cpu_median_us=330
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=400 cpu_median_us=313
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=459 cpu_median_us=26
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=5138 cpu_median_us=3827
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=320 cpu_median_us=56
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=1547 cpu_median_us=2032
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1816 cpu_median_us=611
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1067 cpu_median_us=96
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=4466 cpu_median_us=1068
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=447 cpu_median_us=34
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=5999 cpu_median_us=6116
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=965 cpu_median_us=319
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=268 cpu_median_us=18
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=933 cpu_median_us=71
sink decode frames: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" cases=81 refusals=67 on_gpu=15 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 12 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
#
# Run 9. 2026-11-12, run 8's host and adapters (Linux x86_64; Vulkan on the integrated GPU of PCI
#   vendor 0x1002, device 0x1586, and llvmpipe), the capability corpus alone, after the vectors of
#   the sets trained on openly licensed natural pictures were added (research/trained-sets.md):
#   `cargo test --locked --release --no-default-features --features
#   ops_gpu,video_int,audio_int,splat_int,world_int --test ops_gpu -- --nocapture --test-threads=1`
#   in wai/wai-rs, rustc 1.99.1, on the kernel text of runs 6 to 8. The adapter string the driver
#   reports is left out of each line, as in the runs above.
# Run 9
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
#
# Runs 10 to 12 are of the kernel text before the read forms, by hand on 2026-11-12 through rb, rustc 1.99.1,
#   wgpu and naga 30.0.1, release builds: in wai/wai-ops-gpu, `WAI_OPS_GPU_REQUIRE=1 cargo test
#   --locked --release -- --nocapture --test-threads=1` and `cargo clippy --locked --release
#   --all-targets -- -D warnings`; in wai/wai-rs, under WAI_OPS_GPU_REQUIRE=1, `cargo test --locked
#   --release --no-default-features --features ops_gpu,video_int,audio_int,splat_int,world_int
#   --test ops_gpu` and `cargo test --locked --release --no-default-features --features
#   sink_decode --test sink_decode --test sink_decode_frames`, each with `-- --nocapture
#   --test-threads=1`, and `cargo clippy --locked --release --all-targets --no-default-features
#   --features <set> -- -D warnings` with each of those two feature sets. Every test passed on
#   each. Each clippy
#   run was clean, except that the first run of wai-ops-gpu's on runs 11 and 12 found one lint in
#   a test (a helper of tests/tile.rs taking eight arguments), which this revision allows there;
#   run again on the same kernel text, it was clean on both hosts.
# Run 10. An arm64 laptop's integrated GPU, Metal, macOS 27.2.1 (the laptop of runs 3 and 6).
# Run 11. An arm64 desktop's integrated GPU of the earlier generation of runs 4 and 7, Metal,
#   macOS 27.2.1: a second desktop of that generation (28 logical cores).
# Run 12. A Linux x86_64 host (kernel 7.0.0), Vulkan: the integrated GPU of PCI vendor 0x1002 and
#   llvmpipe (the host of runs 5, 8 and 9).
# Run 10
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.5
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 14 of 14
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2569,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2569 us fills the reader's 1000000 us window only in 390 runs, more than 64"},"cpu":{"median_us":600,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 600 us fills the reader's 1000000 us window only in 1667 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":5391,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 5391 us fills the reader's 1000000 us window only in 186 runs, more than 64"},"cpu":{"median_us":2527,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2527 us fills the reader's 1000000 us window only in 396 runs, more than 64"},"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=0 decodes_on_cpu=16 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=870 cpu_median_us=26
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=892 cpu_median_us=48
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=975 cpu_median_us=43
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=2238 cpu_median_us=212
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=268 cpu_median_us=214
sink decode choice: key=40539b26087ab497 cap=wai.neural.int_hyper unit=cpu gpu_median_us=5609 cpu_median_us=3094
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=830 cpu_median_us=12
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=2462 cpu_median_us=1445
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=643 cpu_median_us=34
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=cpu gpu_median_us=3698 cpu_median_us=1080
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1478 cpu_median_us=433
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=2371 cpu_median_us=792
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1399 cpu_median_us=92
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=1876 cpu_median_us=831
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=1182 cpu_median_us=16
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=cpu gpu_median_us=5383 cpu_median_us=2875
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=981 cpu_median_us=236
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=591 cpu_median_us=15
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=913 cpu_median_us=60
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 11
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.2
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 14 of 14
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2210,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2210 us fills the reader's 1000000 us window only in 453 runs, more than 64"},"cpu":{"median_us":860,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 860 us fills the reader's 1000000 us window only in 1163 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":5440,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 5440 us fills the reader's 1000000 us window only in 184 runs, more than 64"},"cpu":{"median_us":2604,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2604 us fills the reader's 1000000 us window only in 385 runs, more than 64"},"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=0 decodes_on_cpu=16 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1107 cpu_median_us=33
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=1175 cpu_median_us=45
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1269 cpu_median_us=56
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=2735 cpu_median_us=202
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=381 cpu_median_us=229
sink decode choice: key=40539b26087ab497 cap=wai.neural.int_hyper unit=cpu gpu_median_us=6191 cpu_median_us=3322
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=1091 cpu_median_us=24
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=2505 cpu_median_us=1811
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=297 cpu_median_us=36
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=cpu gpu_median_us=3368 cpu_median_us=1070
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1797 cpu_median_us=359
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=2222 cpu_median_us=1036
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1763 cpu_median_us=83
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=2173 cpu_median_us=858
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=1533 cpu_median_us=35
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=cpu gpu_median_us=5470 cpu_median_us=2638
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1267 cpu_median_us=239
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=329 cpu_median_us=18
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=1206 cpu_median_us=56
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 12
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-vulkan-1002 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus narrow: key=swgpu-vulkan-10005 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-vulkan-1002 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=swgpu-vulkan-10005 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=0.6
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.1
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu narrow mutants: key=gpu-vulkan-1002 killed 19 of 19
ops-gpu narrow mutants: key=swgpu-vulkan-10005 killed 19 of 19
ops-gpu narrow: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu narrow: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-vulkan-1002 killed 14 of 14
ops-gpu tile mutants: key=swgpu-vulkan-10005 killed 14 of 14
ops-gpu tile: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu tile: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="RAPL package energy (the whole package: every process on it)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":811,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 811 us fills the reader's 500000 us window only in 617 runs, more than 64"},"cpu":{"median_us":628,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 628 us fills the reader's 500000 us window only in 797 runs, more than 64"},"measured_now":true,"graphs":1},{"key":"df6e4afccedb5dc2","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":3850,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3850 us fills the reader's 500000 us window only in 130 runs, more than 64"},"cpu":{"median_us":3207,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3207 us fills the reader's 500000 us window only in 156 runs, more than 64"},"measured_now":true,"graphs":1}]}
sink decode choice: backend=vulkan shapes=19 decodes_on_gpu=5 decodes_on_cpu=11 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=149 cpu_median_us=18
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=225 cpu_median_us=50
sink decode choice: key=22ae38c965880e1b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=296 cpu_median_us=36
sink decode choice: key=25f9209e6a87532e cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=716 cpu_median_us=286
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=gpu gpu_median_us=176 cpu_median_us=181
sink decode choice: key=40539b26087ab497 cap=wai.neural.int_hyper unit=gpu gpu_median_us=3870 cpu_median_us=4435
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=129 cpu_median_us=5
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=1865 cpu_median_us=2359
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=136 cpu_median_us=25
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=1413 cpu_median_us=1797
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=539 cpu_median_us=450
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=908 cpu_median_us=735
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=805 cpu_median_us=76
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=813 cpu_median_us=654
sink decode choice: key=d8cb91aea662fd5f cap=wai.video.int_hyper unit=cpu gpu_median_us=234 cpu_median_us=18
sink decode choice: key=df6e4afccedb5dc2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=3509 cpu_median_us=4834
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=314 cpu_median_us=252
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=78 cpu_median_us=10
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=272 cpu_median_us=53
sink decode frames: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
#
# Runs 13 to 15 are of this revision's kernel text, the read forms (out.wgsl: rgb8, lows) added, by
#   hand on 2026-11-12 through rb, rustc 1.99.1, wgpu and naga 30.0.1, release builds: the commands
#   of runs 10 to 12 (in wai/wai-ops-gpu, every test of the crate under WAI_OPS_GPU_REQUIRE=1, the
#   new tests/out.rs among them, and clippy with every target; in wai/wai-rs, tests/ops_gpu.rs with
#   ops_gpu,video_int,audio_int,splat_int,world_int, tests/sink_decode.rs and
#   tests/sink_decode_frames.rs with sink_decode, and clippy with every target and each of those two
#   feature sets). Every test passed on each and each clippy run was clean, on the revision these
#   lines were recorded from (on runs 13 and 14, a first run of the clippy with sink_decode found an
#   unused import in tests/sink_decode_speed.rs, removed before the recorded run).
# Run 13. An arm64 laptop's integrated GPU, Metal, macOS 27.2.1 (the laptop of runs 3, 6 and 10).
# Run 14. An arm64 desktop's integrated GPU of the earlier generation, Metal, macOS 27.2.1 (the
#   desktop of run 11).
# Run 15. A Linux x86_64 host (kernel 7.0.0), Vulkan: the integrated GPU of PCI vendor 0x1002 and
#   llvmpipe (the host of runs 5, 8, 9 and 12).
# Run 13
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=1.4
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-metal-0 killed 4 of 4
ops-gpu out: key=gpu-metal-0 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 14 of 14
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2476,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2476 us fills the reader's 1000000 us window only in 404 runs, more than 64"},"cpu":{"median_us":580,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 580 us fills the reader's 1000000 us window only in 1725 runs, more than 64"},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":3469,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3469 us fills the reader's 1000000 us window only in 289 runs, more than 64"},"cpu":{"median_us":2217,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2217 us fills the reader's 1000000 us window only in 452 runs, more than 64"},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":5179},{"kernel":"igdn_r","median_us":3469}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=0 decodes_on_cpu=16 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=825 cpu_median_us=23
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=861 cpu_median_us=51
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=cpu gpu_median_us=3553 cpu_median_us=2388
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=248 cpu_median_us=216
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=833 cpu_median_us=10
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=2437 cpu_median_us=1401
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=927 cpu_median_us=38
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=239 cpu_median_us=33
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=915 cpu_median_us=20
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=cpu gpu_median_us=2402 cpu_median_us=1124
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1071 cpu_median_us=360
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=1850 cpu_median_us=784
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1376 cpu_median_us=94
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=1762 cpu_median_us=645
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=cpu gpu_median_us=3775 cpu_median_us=2873
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=937 cpu_median_us=227
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1582 cpu_median_us=253
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=210 cpu_median_us=16
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=908 cpu_median_us=60
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 14
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.1
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-metal-0 killed 4 of 4
ops-gpu out: key=gpu-metal-0 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 14 of 14
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2325,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2325 us fills the reader's 1000000 us window only in 431 runs, more than 64"},"cpu":{"median_us":827,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 827 us fills the reader's 1000000 us window only in 1210 runs, more than 64"},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":4211,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 4211 us fills the reader's 1000000 us window only in 238 runs, more than 64"},"cpu":{"median_us":3202,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3202 us fills the reader's 1000000 us window only in 313 runs, more than 64"},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":5501},{"kernel":"igdn_r","median_us":4211}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=0 decodes_on_cpu=16 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1051 cpu_median_us=34
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=1177 cpu_median_us=44
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=cpu gpu_median_us=4177 cpu_median_us=3673
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=861 cpu_median_us=201
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=1045 cpu_median_us=10
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=2496 cpu_median_us=2479
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1219 cpu_median_us=44
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=902 cpu_median_us=35
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=1177 cpu_median_us=28
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=cpu gpu_median_us=2237 cpu_median_us=1030
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1429 cpu_median_us=354
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=2240 cpu_median_us=1008
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1741 cpu_median_us=86
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=2161 cpu_median_us=838
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=cpu gpu_median_us=4384 cpu_median_us=4029
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1249 cpu_median_us=231
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=2020 cpu_median_us=223
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=879 cpu_median_us=17
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=1170 cpu_median_us=55
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 15
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-vulkan-1002 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus narrow: key=swgpu-vulkan-10005 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-vulkan-1002 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=swgpu-vulkan-10005 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.6
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=12.7
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu narrow mutants: key=gpu-vulkan-1002 killed 19 of 19
ops-gpu narrow mutants: key=swgpu-vulkan-10005 killed 19 of 19
ops-gpu narrow: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu narrow: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-vulkan-1002 killed 4 of 4
ops-gpu out mutants: key=swgpu-vulkan-10005 killed 4 of 4
ops-gpu out: key=gpu-vulkan-1002 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu out: key=swgpu-vulkan-10005 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-vulkan-1002 killed 14 of 14
ops-gpu tile mutants: key=swgpu-vulkan-10005 killed 14 of 14
ops-gpu tile: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu tile: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=185 decodes=902 tiled=155 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
sink decode choice energy: reader="RAPL package energy (the whole package: every process on it)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"gpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":4082,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 4082 us fills the reader's 500000 us window only in 123 runs, more than 64"},"cpu":{"median_us":11923,"runs":5,"energy_uj":2259203,"energy_runs":42,"no_energy":null},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"gpu","by":"energy","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":12599,"runs":5,"energy_uj":1028133,"energy_runs":40,"no_energy":null},"cpu":{"median_us":47867,"runs":5,"energy_uj":20257527,"energy_runs":11,"no_energy":null},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":12803},{"kernel":"igdn_r","median_us":12599}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=vulkan shapes=19 decodes_on_gpu=5 decodes_on_cpu=11 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=498 cpu_median_us=33
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=718 cpu_median_us=77
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=gpu gpu_median_us=12562 cpu_median_us=47209
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=1677 cpu_median_us=312
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=438 cpu_median_us=9
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=4929 cpu_median_us=36729
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1143 cpu_median_us=65
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=1198 cpu_median_us=55
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=759 cpu_median_us=38
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=2361 cpu_median_us=3921
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=gpu gpu_median_us=1958 cpu_median_us=4951
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=gpu gpu_median_us=3713 cpu_median_us=7015
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=3558 cpu_median_us=122
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=gpu gpu_median_us=3961 cpu_median_us=6447
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=12713 cpu_median_us=272688
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1045 cpu_median_us=914
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=2900 cpu_median_us=421
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=1311 cpu_median_us=25
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=844 cpu_median_us=73
sink decode frames: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
#
# Runs 16 to 18 are of this revision's kernel text, the tiled convolution's bound taken per tile and
#   stage (tile.wgsl: mag_tile, and each stage of conv_tile under its own chunk, folding a row of taps
#   at a time), by hand on 2026-11-12 through rb, rustc 1.99.1, wgpu and naga 30.0.1, release builds:
#   the commands of runs 13 to 15 (in wai/wai-ops-gpu, every test of the crate under
#   WAI_OPS_GPU_REQUIRE=1, the new tests/timing.rs among them, and clippy with every target; in
#   wai/wai-rs, tests/ops_gpu.rs with ops_gpu,video_int,audio_int,splat_int,world_int,
#   tests/sink_decode.rs and tests/sink_decode_frames.rs with sink_decode, and clippy with every
#   target and each of those two feature sets). Every test passed on each and each clippy run was
#   clean, on the revision these lines were recorded from, except on run 18: there the first clippy
#   run of wai-ops-gpu found one lint (exec.rs's stamps_after read its eight-byte words with
#   `chunks_exact`), rewritten with `as_chunks` before runs 16 and 17, after which clippy and
#   tests/timing.rs ran again on that host, clean and passing; and its timing test then held the two
#   profiled dispatches to their entry points' names, where it now holds them to their count. On run
#   17 the host's memory cap (12 GB) stopped the build of tests/sink_decode.rs; that command and the two
#   clippy runs ran again on the same revision with four build jobs, and the run's lines are both
#   invocations' reports.
# Run 16. An arm64 laptop's integrated GPU, Metal, macOS 27.2.1 (the laptop of runs 3, 6, 10 and 13).
# Run 17. An arm64 desktop's integrated GPU of the earlier generation, Metal, macOS 27.2.1 (the
#   desktop of runs 11 and 14).
# Run 18. A Linux x86_64 host (kernel 7.0.0), Vulkan: the integrated GPU of PCI vendor 0x1002 and
#   llvmpipe (the host of runs 5, 8, 9, 12 and 15).
# Run 16
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.3
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-metal-0 killed 4 of 4
ops-gpu out: key=gpu-metal-0 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 20 of 20
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 decodes=1159 tiled=162 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu timing: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" runs=4 device_us=[Some(1993), Some(1982), Some(1980), Some(1982)] profile_us=[("mag_tile", 87), ("conv_tile", 1897)]
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2295,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2295 us fills the reader's 1000000 us window only in 436 runs, more than 64"},"cpu":{"median_us":1572,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1572 us fills the reader's 1000000 us window only in 637 runs, more than 64"},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"gpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":6190,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 6190 us fills the reader's 1000000 us window only in 162 runs, more than 64"},"cpu":{"median_us":6899,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 6899 us fills the reader's 1000000 us window only in 145 runs, more than 64"},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":7281},{"kernel":"igdn_r","median_us":6190}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=4 decodes_on_cpu=12 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1552 cpu_median_us=73
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=1349 cpu_median_us=132
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=gpu gpu_median_us=3241 cpu_median_us=7282
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=800 cpu_median_us=498
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=1118 cpu_median_us=21
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=5643 cpu_median_us=7608
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1200 cpu_median_us=114
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=748 cpu_median_us=82
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=1181 cpu_median_us=43
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=2975 cpu_median_us=3028
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1423 cpu_median_us=682
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=2710 cpu_median_us=1659
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1244 cpu_median_us=242
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=2768 cpu_median_us=1874
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=3723 cpu_median_us=9523
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1327 cpu_median_us=483
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1874 cpu_median_us=643
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=360 cpu_median_us=39
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=1215 cpu_median_us=158
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 17
ops-gpu arith: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-metal-0 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-metal-0 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=2.2
ops-gpu mutants: key=gpu-metal-0 killed 18 of 18
ops-gpu narrow mutants: key=gpu-metal-0 killed 19 of 19
ops-gpu narrow: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-metal-0 killed 4 of 4
ops-gpu out: key=gpu-metal-0 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-metal-0 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-metal-0 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-metal-0 killed 20 of 20
ops-gpu tile: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" cases=192 decodes=1159 tiled=162 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu timing: key=gpu-metal-0 backend=metal type=integrated vendor=0x0 device=0x0 driver="" driver_info="" runs=4 device_us=[Some(699), Some(701), Some(810), Some(811)] profile_us=[("mag_tile", 46), ("conv_tile", 769)]
sink decode choice energy: reader="this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":2491,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2491 us fills the reader's 1000000 us window only in 402 runs, more than 64"},"cpu":{"median_us":970,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 970 us fills the reader's 1000000 us window only in 1031 runs, more than 64"},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"cpu","by":"time","trigger":"first","reader":"this process's CPU energy and its GPU energy (IOReport GPU Energy, the whole rail where this process was the GPU's sole user)","gpu":{"median_us":3209,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3209 us fills the reader's 1000000 us window only in 312 runs, more than 64"},"cpu":{"median_us":2885,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 2885 us fills the reader's 1000000 us window only in 347 runs, more than 64"},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":4402},{"kernel":"igdn_r","median_us":3209}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=metal shapes=19 decodes_on_gpu=0 decodes_on_cpu=16 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1165 cpu_median_us=36
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=1266 cpu_median_us=54
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=cpu gpu_median_us=3183 cpu_median_us=2838
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=397 cpu_median_us=229
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=1177 cpu_median_us=12
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=cpu gpu_median_us=2837 cpu_median_us=1854
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1319 cpu_median_us=48
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=810 cpu_median_us=36
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=1271 cpu_median_us=20
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=cpu gpu_median_us=2439 cpu_median_us=1230
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1504 cpu_median_us=449
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=2422 cpu_median_us=1131
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1324 cpu_median_us=99
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=2383 cpu_median_us=933
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=cpu gpu_median_us=3487 cpu_median_us=3470
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1357 cpu_median_us=287
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1996 cpu_median_us=252
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=337 cpu_median_us=16
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=1280 cpu_median_us=63
sink decode frames: backend=metal type=integrated vendor=0x0 device=0x0 driver="" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=metal type=integrated vendor=0x0 device=0x0 driver="" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers
# Run 18
ops-gpu arith: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=1051757 words=42070280 equal
ops-gpu arith: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=1051757 words=42070280 equal
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.audio.int_codec cases=15 refusals=14 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_hyper cases=15 refusals=12 graphs_gpu_acc32=0 graphs_gpu_acc64=8 graphs_cpu=0 [] graphs_compared=8 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_mlicpp cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=20 graphs_cpu=0 [] graphs_compared=20 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.neural.int_synth cases=4 refusals=0 graphs_gpu_acc32=3 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.splat.int_codec cases=12 refusals=11 graphs_gpu_acc32=0 graphs_gpu_acc64=1 graphs_cpu=0 [] graphs_compared=1 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_hyper cases=22 refusals=21 graphs_gpu_acc32=0 graphs_gpu_acc64=4 graphs_cpu=0 [] graphs_compared=4 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.video.int_motion cases=11 refusals=9 graphs_gpu_acc32=0 graphs_gpu_acc64=0 graphs_cpu=0 [] graphs_compared=0 graphs_differ=0 differ=0 errors=0
ops-gpu capability: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cap=wai.world.model cases=2 refusals=0 graphs_gpu_acc32=0 graphs_gpu_acc64=34 graphs_cpu=0 [] graphs_compared=34 graphs_differ=0 differ=0 errors=0
ops-gpu capability: wai.image.jpegai NOT RUN: WAI_JPEGAI_CODESTREAM, WAI_JPEGAI_WEIGHTS_DIR and WAI_JPEGAI_NORM are not all set (tools/wai_jpegai_fetch_e2e_inputs.sh)
ops-gpu corpus narrow: key=gpu-vulkan-1002 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus narrow: key=swgpu-vulkan-10005 narrow_ops=293 narrow_vectors=181; with narrowing off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=gpu-vulkan-1002 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus tiled: key=swgpu-vulkan-10005 tiled_ops=362 tiled_vectors=197; with tiling off: acc64 vectors=1426 decodes=1921
ops-gpu corpus: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=3.5
ops-gpu corpus: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" acc32 vectors=813 decodes=1104; acc64 vectors=1426 decodes=1921; refusals=212; cpu=209 load_refused=95 mismatches=0 errors=0 secs=9.5
ops-gpu mutants: key=gpu-vulkan-1002 killed 18 of 18
ops-gpu mutants: key=swgpu-vulkan-10005 killed 18 of 18
ops-gpu narrow mutants: key=gpu-vulkan-1002 killed 19 of 19
ops-gpu narrow mutants: key=swgpu-vulkan-10005 killed 19 of 19
ops-gpu narrow: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu narrow: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=192 edge=18 runtime=14 decodes=962 tiled_mismatches=0 narrow_ops=155 finite_chunks=73 one_channel_mismatches=0 narrow_mismatches=0 wide_mismatches=0
ops-gpu out mutants: key=gpu-vulkan-1002 killed 4 of 4
ops-gpu out mutants: key=swgpu-vulkan-10005 killed 4 of 4
ops-gpu out: key=gpu-vulkan-1002 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu out: key=swgpu-vulkan-10005 kernels=shipped rgb8_decodes=56 on_device=56 refused=6 part_reads=16 resident_writes=16 igdn_runs=9 differ=0 equal=true
ops-gpu semantics mutant: key=gpu-vulkan-1002 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics mutant: key=swgpu-vulkan-10005 refused: the WGSL semantics suite: rem_i32 (ia % ib) a=0x00000001 b=0x00000000 c=0x00000000 d=0x00000000: WGSL gives 0x00000000, got 0x00000001
ops-gpu semantics: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu semantics: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" committed=25088 cases=254464 guarded=231690 differ=0 unguarded=22774 unguarded_differ=0 unguarded_ops=none
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=gpu-vulkan-1002 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_hyper payloads=6 refused=1 graphs_gpu=10 graphs_compared=10 graphs_differ=0 errors=0 equal=true
ops-gpu sequence: key=swgpu-vulkan-10005 cap=wai.neural.int_mlicpp payloads=6 refused=1 graphs_gpu=50 graphs_compared=50 graphs_differ=0 errors=0 equal=true
ops-gpu tile mutants: key=gpu-vulkan-1002 killed 20 of 20
ops-gpu tile mutants: key=swgpu-vulkan-10005 killed 20 of 20
ops-gpu tile: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" cases=192 decodes=1159 tiled=162 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu tile: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" cases=192 decodes=1159 tiled=162 fused=56 fused_igdn=21 igdn_fused_mismatches=0 unfused_mismatches=0 tiled_mismatches=0 untiled_mismatches=0
ops-gpu timing: key=gpu-vulkan-1002 backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" driver_info="Mesa 26.0.8-1ubuntu0.3" runs=4 device_us=[Some(1084), Some(469), Some(465), Some(466)] profile_us=[("mag_tile", 60), ("conv_tile", 403)]
ops-gpu timing: key=swgpu-vulkan-10005 backend=vulkan type=cpu vendor=0x10005 device=0x0 driver="llvmpipe" driver_info="Mesa 26.0.8-1ubuntu0.3 (LLVM 21.1.8)" runs=4 device_us=[Some(26820), Some(17849), Some(18341), Some(14605)] profile_us=[("mag_tile", 993), ("conv_tile", 11434)]
sink decode choice energy: reader="RAPL package energy (the whole package: every process on it)" vector=leaves choice={"rule":"energy where both units are read, time otherwise","graphs":[{"key":"cf43be49906f0180","ops":5,"unit":"cpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":3477,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 3477 us fills the reader's 500000 us window only in 144 runs, more than 64"},"cpu":{"median_us":1278,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 1278 us fills the reader's 500000 us window only in 392 runs, more than 64"},"kernel":null,"kernels":[],"measured_now":true,"graphs":1},{"key":"208af0953f94da47","ops":9,"unit":"gpu","by":"time","trigger":"first","reader":"RAPL package energy (the whole package: every process on it)","gpu":{"median_us":5203,"runs":5,"energy_uj":null,"energy_runs":0,"no_energy":"a run of 5203 us fills the reader's 500000 us window only in 97 runs, more than 64"},"cpu":{"median_us":11801,"runs":5,"energy_uj":687664,"energy_runs":43,"no_energy":null},"kernel":"igdn_r","kernels":[{"kernel":"igdn_tile","median_us":5706},{"kernel":"igdn_r","median_us":5203}],"measured_now":true,"graphs":1}]}
sink decode choice: backend=vulkan shapes=19 decodes_on_gpu=4 decodes_on_cpu=12 mismatches=0
sink decode choice: key=08c8797c73c25ec0 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=569 cpu_median_us=30
sink decode choice: key=1bbd889a55c9753c cap=wai.video.int_cond unit=cpu gpu_median_us=775 cpu_median_us=61
sink decode choice: key=208af0953f94da47 cap=wai.neural.int_hyper unit=gpu gpu_median_us=5196 cpu_median_us=5914
sink decode choice: key=390a7f75fbc89437 cap=wai.neural.int_synth unit=cpu gpu_median_us=415 cpu_median_us=276
sink decode choice: key=579973a919e9ba7e cap=wai.neural.int_hyper unit=cpu gpu_median_us=520 cpu_median_us=13
sink decode choice: key=5a9235945455f20f cap=wai.neural.int_synth unit=gpu gpu_median_us=3910 cpu_median_us=4998
sink decode choice: key=6394b6a0dc33951b cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=1113 cpu_median_us=66
sink decode choice: key=781109cf66ca73ea cap=wai.neural.int_synth unit=cpu gpu_median_us=259 cpu_median_us=43
sink decode choice: key=9ee6117cd3b09cf5 cap=wai.video.int_hyper unit=cpu gpu_median_us=892 cpu_median_us=44
sink decode choice: key=a5fe4a6238128211 cap=wai.audio.int_codec unit=gpu gpu_median_us=2422 cpu_median_us=2804
sink decode choice: key=ad727048a1cffac1 cap=wai.splat.int_codec unit=cpu gpu_median_us=1737 cpu_median_us=1134
sink decode choice: key=b1b1300cd59690fd cap=wai.neural.int_hyper unit=cpu gpu_median_us=3607 cpu_median_us=1124
sink decode choice: key=ccc55b80400c82f2 cap=wai.video.int_cond unit=cpu gpu_median_us=1263 cpu_median_us=118
sink decode choice: key=cf43be49906f0180 cap=wai.neural.int_hyper unit=cpu gpu_median_us=3380 cpu_median_us=999
sink decode choice: key=d3da60b30772fea2 cap=wai.neural.int_hyper unit=gpu gpu_median_us=6285 cpu_median_us=7721
sink decode choice: key=e0012147b3c9910b cap=wai.video.int_cond unit=cpu gpu_median_us=1273 cpu_median_us=338
sink decode choice: key=e7e850bb240e5d01 cap=wai.neural.int_mlicpp unit=cpu gpu_median_us=2038 cpu_median_us=392
sink decode choice: key=efc423150f37c4f7 cap=wai.neural.int_synth unit=cpu gpu_median_us=266 cpu_median_us=16
sink decode choice: key=fa7d02f9c360212a cap=wai.video.int_cond unit=cpu gpu_median_us=891 cpu_median_us=66
sink decode frames: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" streams=21 objects=95 frames=70 refusals=25 on_gpu=26 graphs_compared=98 graphs_differ=0 mismatches=0
sink decode frames: the C interface: 3 frame forms, 16 objects, the Rust API's bytes and answers
sink decode: backend=vulkan type=integrated vendor=0x1002 device=0x1586 driver="radv" cases=82 refusals=67 on_gpu=16 on_cpu=66 capabilities=8 mismatches=0
sink decode: planted "a floating-point partial product in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a fold zero-extended in the tiled kernel": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "a narrow chunk zero-extended in the four-channel kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 32-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "rounding shift one short in the 64-bit kernels": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's clamp without its top bound in the picture's bytes (a clamp)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the RGB8 step's rescale rounded down in the picture's bytes (a rounding mode)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the low-word read's second range one element on": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: planted "the picture's channels written BGR for RGB (a plane order)": caught at start-up; 13 decodes on the CPU, each reporting dispatch_unit_inexact
sink decode: the C interface: 8 capabilities, the Rust API's bytes and answers