Skip to content

Remove _CCCL_GRID_CONSTANT from scan kernel - #9854

Closed
bernhardmgruber wants to merge 1 commit into
NVIDIA:mainfrom
bernhardmgruber:fix_scan_reg
Closed

bernhardmgruber wants to merge 1 commit into
NVIDIA:mainfrom
bernhardmgruber:fix_scan_reg

Conversation

@bernhardmgruber

Copy link
Copy Markdown
Contributor

Fixes: #9851

@bernhardmgruber
bernhardmgruber requested a review from a team as a code owner July 14, 2026 11:54
@github-project-automation github-project-automation Bot moved this to Todo in CCCL Jul 14, 2026
@cccl-authenticator-app cccl-authenticator-app Bot moved this from Todo to In Review in CCCL Jul 14, 2026
@coderabbitai

coderabbitai Bot commented Jul 14, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: bbc1f1c8-138e-423e-92bb-428a376fca36

📥 Commits

Reviewing files that changed from the base of the PR and between 22e4e68 and f29ec53.

📒 Files selected for processing (1)
  • cub/cub/device/dispatch/kernels/kernel_scan.cuh

📝 Walkthrough

Summary by CodeRabbit

  • Bug Fixes
    • Improved compatibility and reliability for device-wide scan operations across supported CUDA environments.
    • Standardized scan kernel argument handling to prevent compilation or execution issues related to parameter qualifiers.

Walkthrough

The scan initialization and execution kernel signatures in kernel_scan.cuh remove _CCCL_GRID_CONSTANT qualifiers from their parameters while retaining plain const declarations.

Changes

Scan kernel signature updates

Layer / File(s) Summary
Update scan kernel signatures
cub/cub/device/dispatch/kernels/kernel_scan.cuh
DeviceScanInitKernel and DeviceScanKernel remove _CCCL_GRID_CONSTANT from their parameters, including the CUDA-version-conditional qualifier around init_value.

Assessment against linked issues

Objective Addressed Explanation
Fix the scan_exclusive_sum regression in CCCL 3.4 on Ampere [#9851] The signature changes appear related, but the provided changes do not include a test or explicit Ampere-specific validation confirming the regression is fixed.

Possibly related PRs

  • NVIDIA/cccl#9829: Removes the same kernel parameter qualifier from a related CUB dispatch kernel.
  • NVIDIA/cccl#9852: Applies the same _CCCL_GRID_CONSTANT removal pattern to related dispatch kernels.

Suggested reviewers: srinivasyadav18, davebayer, griwes


Comment @coderabbitai help to get the list of available commands.

@elstehle

Copy link
Copy Markdown
Contributor

#9851 lists the Ampere regressions/fix. How does this PR influence behaviour on other architectures, e.g., look-ahead-enabled ones? Asking, because, I've previously seen both directions with _CCCL_GRID_CONSTANT annotations added/removed.

@bernhardmgruber

Copy link
Copy Markdown
Contributor Author

That's a fair point. I only benchmarked that this PR mitigates the regression on Ampere. I did not check the impact on other architectures.

Here is CCCL 3.4 vs this PR on A100:

## [0] NVIDIA A100 80GB PCIe

|  T{ct}  |  OffsetT{ct}  |  Elements{io}  |   Ref Time |   Ref Noise |   Cmp Time |   Cmp Noise |       Diff |   %Diff |  Status  |
|---------|---------------|----------------|------------|-------------|------------|-------------|------------|---------|----------|
|   I8    |      I32      |      2^16      |  10.340 us |       3.97% |  10.322 us |       4.11% |  -0.019 us |  -0.18% |  🔵 SAME  |
|   I8    |      I32      |      2^20      |  15.418 us |      14.77% |  15.310 us |       8.17% |  -0.108 us |  -0.70% |  🔵 SAME  |
|   I8    |      I32      |      2^24      |  69.668 us |       1.99% |  69.675 us |       2.08% |   0.007 us |   0.01% |  🔵 SAME  |
|   I8    |      I32      |      2^28      | 936.423 us |       0.31% | 938.353 us |       0.32% |   1.930 us |   0.21% |  🔵 SAME  |
|   I8    |      I64      |      2^16      |  10.625 us |       4.99% |  10.544 us |       4.60% |  -0.081 us |  -0.77% |  🔵 SAME  |
|   I8    |      I64      |      2^20      |  15.669 us |      20.31% |  15.388 us |      12.84% |  -0.281 us |  -1.79% |  🔵 SAME  |
|   I8    |      I64      |      2^24      |  81.028 us |       1.20% |  81.214 us |       1.23% |   0.186 us |   0.23% |  🔵 SAME  |
|   I8    |      I64      |      2^28      |   1.128 ms |       0.25% |   1.129 ms |       0.22% |   0.742 us |   0.07% |  🔵 SAME  |
|   I8    |      I64      |      2^32      |  17.893 ms |       0.16% |  17.906 ms |       0.17% |  12.848 us |   0.07% |  🔵 SAME  |
|   I16   |      I32      |      2^16      |  11.027 us |       4.41% |  11.025 us |       4.52% |  -0.001 us |  -0.01% |  🔵 SAME  |
|   I16   |      I32      |      2^20      |  16.985 us |      11.30% |  17.089 us |      10.94% |   0.104 us |   0.61% |  🔵 SAME  |
|   I16   |      I32      |      2^24      |  87.541 us |       1.21% |  87.598 us |       1.18% |   0.057 us |   0.06% |  🔵 SAME  |
|   I16   |      I32      |      2^28      |   1.181 ms |       0.32% |   1.183 ms |       0.32% |   1.838 us |   0.16% |  🔵 SAME  |
|   I16   |      I64      |      2^16      |  10.884 us |       5.16% |  10.887 us |       4.79% |   0.002 us |   0.02% |  🔵 SAME  |
|   I16   |      I64      |      2^20      |  15.509 us |       2.94% |  15.561 us |       2.83% |   0.052 us |   0.33% |  🔵 SAME  |
|   I16   |      I64      |      2^24      |  85.391 us |       1.45% |  85.566 us |       1.63% |   0.175 us |   0.21% |  🔵 SAME  |
|   I16   |      I64      |      2^28      |   1.183 ms |       0.29% |   1.179 ms |       0.31% |  -3.791 us |  -0.32% |  🟢 FAST  |
|   I16   |      I64      |      2^32      |  18.777 ms |       0.16% |  18.723 ms |       0.16% | -54.027 us |  -0.29% |  🟢 FAST  |
|   I32   |      I32      |      2^16      |  10.971 us |       4.40% |  10.947 us |       4.72% |  -0.024 us |  -0.22% |  🔵 SAME  |
|   I32   |      I32      |      2^20      |  18.291 us |       3.31% |  18.340 us |       3.00% |   0.049 us |   0.27% |  🔵 SAME  |
|   I32   |      I32      |      2^24      | 108.340 us |       1.99% | 108.534 us |       1.95% |   0.194 us |   0.18% |  🔵 SAME  |
|   I32   |      I32      |      2^28      |   1.514 ms |       1.04% |   1.514 ms |       1.11% |  -0.026 us |  -0.00% |  🔵 SAME  |
|   I32   |      I64      |      2^16      |  11.044 us |       4.23% |  10.969 us |       4.55% |  -0.075 us |  -0.68% |  🔵 SAME  |
|   I32   |      I64      |      2^20      |  17.855 us |       3.81% |  18.065 us |       4.63% |   0.209 us |   1.17% |  🔵 SAME  |
|   I32   |      I64      |      2^24      | 105.971 us |       2.12% | 105.965 us |       2.16% |  -0.005 us |  -0.01% |  🔵 SAME  |
|   I32   |      I64      |      2^28      |   1.505 ms |       0.68% |   1.505 ms |       0.72% |  -0.101 us |  -0.01% |  🔵 SAME  |
|   I32   |      I64      |      2^32      |  24.170 ms |       0.58% |  24.165 ms |       0.55% |  -4.177 us |  -0.02% |  🔵 SAME  |
|   I64   |      I32      |      2^16      |  11.882 us |       4.33% |  11.913 us |       4.28% |   0.031 us |   0.26% |  🔵 SAME  |
|   I64   |      I32      |      2^20      |  24.130 us |       8.51% |  24.406 us |      11.99% |   0.276 us |   1.14% |  🔵 SAME  |
|   I64   |      I32      |      2^24      | 190.874 us |       1.38% | 190.979 us |       1.43% |   0.106 us |   0.06% |  🔵 SAME  |
|   I64   |      I32      |      2^28      |   2.795 ms |       1.22% |   2.794 ms |       1.23% |  -0.981 us |  -0.04% |  🔵 SAME  |
|   I64   |      I64      |      2^16      |  11.835 us |       4.41% |  11.824 us |       4.52% |  -0.011 us |  -0.09% |  🔵 SAME  |
|   I64   |      I64      |      2^20      |  24.362 us |      10.68% |  23.795 us |       3.85% |  -0.568 us |  -2.33% |  🔵 SAME  |
|   I64   |      I64      |      2^24      | 188.281 us |       1.86% | 187.242 us |       1.68% |  -1.040 us |  -0.55% |  🔵 SAME  |
|   I64   |      I64      |      2^28      |   2.751 ms |       0.60% |   2.749 ms |       0.66% |  -1.744 us |  -0.06% |  🔵 SAME  |
|   I64   |      I64      |      2^32      |  44.103 ms |       0.13% |  44.065 ms |       0.11% | -37.969 us |  -0.09% |  🔵 SAME  |
|  I128   |      I32      |      2^16      |  19.217 us |       2.48% |  19.116 us |       2.74% |  -0.102 us |  -0.53% |  🔵 SAME  |
|  I128   |      I32      |      2^20      |  58.094 us |       1.66% |  58.391 us |       1.49% |   0.297 us |   0.51% |  🔵 SAME  |
|  I128   |      I32      |      2^24      | 641.576 us |       0.18% | 642.476 us |       0.18% |   0.899 us |   0.14% |  🔵 SAME  |
|  I128   |      I32      |      2^28      |   9.935 ms |       0.06% |   9.941 ms |       0.06% |   6.121 us |   0.06% |  🔴 SLOW  |
|  I128   |      I64      |      2^16      |  19.036 us |       2.69% |  19.025 us |       2.77% |  -0.010 us |  -0.05% |  🔵 SAME  |
|  I128   |      I64      |      2^20      |  58.401 us |       1.78% |  58.068 us |       1.73% |  -0.333 us |  -0.57% |  🔵 SAME  |
|  I128   |      I64      |      2^24      | 641.341 us |       0.19% | 639.172 us |       0.25% |  -2.169 us |  -0.34% |  🟢 FAST  |
|  I128   |      I64      |      2^28      |   9.938 ms |       0.06% |   9.917 ms |       0.09% | -21.314 us |  -0.21% |  🟢 FAST  |
|   F32   |      I32      |      2^16      |  10.900 us |       4.57% |  10.981 us |       4.48% |   0.081 us |   0.74% |  🔵 SAME  |
|   F32   |      I32      |      2^20      |  18.790 us |       3.77% |  18.611 us |       3.45% |  -0.179 us |  -0.95% |  🔵 SAME  |
|   F32   |      I32      |      2^24      | 110.930 us |       1.73% | 110.811 us |       1.79% |  -0.118 us |  -0.11% |  🔵 SAME  |
|   F32   |      I32      |      2^28      |   1.604 ms |       0.33% |   1.602 ms |       0.36% |  -2.248 us |  -0.14% |  🔵 SAME  |
|   F32   |      I64      |      2^16      |  10.971 us |       4.44% |  10.945 us |       4.64% |  -0.027 us |  -0.24% |  🔵 SAME  |
|   F32   |      I64      |      2^20      |  18.073 us |       3.21% |  18.244 us |       3.03% |   0.171 us |   0.95% |  🔵 SAME  |
|   F32   |      I64      |      2^24      | 111.127 us |       1.84% | 110.716 us |       1.81% |  -0.411 us |  -0.37% |  🔵 SAME  |
|   F32   |      I64      |      2^28      |   1.626 ms |       0.38% |   1.623 ms |       0.37% |  -2.993 us |  -0.18% |  🔵 SAME  |
|   F32   |      I64      |      2^32      |  25.467 ms |       0.50% |  25.416 ms |       0.46% | -50.732 us |  -0.20% |  🔵 SAME  |
|   F64   |      I32      |      2^16      |  11.756 us |       4.42% |  11.725 us |       4.35% |  -0.031 us |  -0.26% |  🔵 SAME  |
|   F64   |      I32      |      2^20      |  26.534 us |      13.72% |  27.066 us |      17.79% |   0.533 us |   2.01% |  🔵 SAME  |
|   F64   |      I32      |      2^24      | 196.332 us |       1.35% | 196.381 us |       1.34% |   0.049 us |   0.02% |  🔵 SAME  |
|   F64   |      I32      |      2^28      |   2.878 ms |       1.04% |   2.879 ms |       1.03% |   1.535 us |   0.05% |  🔵 SAME  |
|   F64   |      I64      |      2^16      |  11.522 us |       3.86% |  11.615 us |       4.31% |   0.094 us |   0.81% |  🔵 SAME  |
|   F64   |      I64      |      2^20      |  28.023 us |      21.94% |  29.263 us |      25.30% |   1.240 us |   4.42% |  🔵 SAME  |
|   F64   |      I64      |      2^24      | 193.241 us |       1.51% | 193.497 us |       1.65% |   0.256 us |   0.13% |  🔵 SAME  |
|   F64   |      I64      |      2^28      |   2.860 ms |       0.51% |   2.861 ms |       0.47% |   0.643 us |   0.02% |  🔵 SAME  |
|   F64   |      I64      |      2^32      |  45.844 ms |       0.06% |  45.864 ms |       0.06% |  20.060 us |   0.04% |  🔵 SAME  |
|   C32   |      I32      |      2^16      |  11.401 us |       3.99% |  11.317 us |       3.33% |  -0.084 us |  -0.74% |  🔵 SAME  |
|   C32   |      I32      |      2^20      |  23.996 us |       2.28% |  24.020 us |       2.24% |   0.025 us |   0.10% |  🔵 SAME  |
|   C32   |      I32      |      2^24      | 196.572 us |       1.27% | 196.460 us |       1.31% |  -0.113 us |  -0.06% |  🔵 SAME  |
|   C32   |      I32      |      2^28      |   2.968 ms |       0.34% |   2.966 ms |       0.32% |  -1.606 us |  -0.05% |  🔵 SAME  |
|   C32   |      I64      |      2^16      |  11.579 us |       4.14% |  11.584 us |       4.14% |   0.004 us |   0.04% |  🔵 SAME  |
|   C32   |      I64      |      2^20      |  24.047 us |       2.27% |  24.180 us |       2.20% |   0.132 us |   0.55% |  🔵 SAME  |
|   C32   |      I64      |      2^24      | 197.551 us |       1.32% | 197.549 us |       1.33% |  -0.003 us |  -0.00% |  🔵 SAME  |
|   C32   |      I64      |      2^28      |   2.936 ms |       0.76% |   2.924 ms |       0.52% | -11.454 us |  -0.39% |  🔵 SAME  |
|   C32   |      I64      |      2^32      |  46.664 ms |       0.07% |  46.649 ms |       0.08% | -14.961 us |  -0.03% |  🔵 SAME  |

@github-actions

Copy link
Copy Markdown
Contributor

😬 CI Workflow Results

🟥 Finished in 2h 30m: Pass: 99%/287 | Total: 12d 06h | Max: 2h 29m | Hits: 19%/1025041

See results here.

@gonidelis

Copy link
Copy Markdown
Member

this pr against main does not regress on b200

(cccl) ggonidelis@umb-b200-261:/src/cccl-pr9854-base/build$ python3 /src/cccl-34x/build/_deps/nvbench-src/python/scripts/nvbench_compare.py --no-color   /src/ses_9854base.json /src/ses_9854.json
['/src/ses_9854base.json', '/src/ses_9854.json']
# base

## [0] NVIDIA B200

|  T{ct}  |  OffsetT{ct}  |  Elements{io}  |   Ref Time |   Ref Noise |   Cmp Time |   Cmp Noise |       Diff |   %Diff |  Status  |
|---------|---------------|----------------|------------|-------------|------------|-------------|------------|---------|----------|
|   I8    |      I32      |      2^16      |  11.254 us |       0.45% |  11.261 us |       0.46% |   0.007 us |   0.07% |  🔵 SAME  |
|   I8    |      I32      |      2^20      |  12.593 us |       7.73% |  12.723 us |       7.15% |   0.130 us |   1.03% |  🔵 SAME  |
|   I8    |      I32      |      2^24      |  27.947 us |       2.63% |  27.908 us |       2.49% |  -0.039 us |  -0.14% |  🔵 SAME  |
|   I8    |      I32      |      2^28      | 209.738 us |       0.51% | 209.899 us |       0.47% |   0.161 us |   0.08% |  🔵 SAME  |
|   I8    |      I64      |      2^16      |  11.262 us |       0.43% |  11.262 us |       0.50% |  -0.001 us |  -0.01% |  🔵 SAME  |
|   I8    |      I64      |      2^20      |  13.210 us |       2.91% |  13.224 us |       2.83% |   0.014 us |   0.11% |  🔵 SAME  |
|   I8    |      I64      |      2^24      |  28.029 us |       2.87% |  27.888 us |       2.47% |  -0.141 us |  -0.50% |  🔵 SAME  |
|   I8    |      I64      |      2^28      | 211.224 us |       0.47% | 211.048 us |       0.49% |  -0.176 us |  -0.08% |  🔵 SAME  |
|   I8    |      I64      |      2^32      |   3.175 ms |       0.27% |   3.173 ms |       0.06% |  -1.119 us |  -0.04% |  🔵 SAME  |
|   I16   |      I32      |      2^16      |  11.265 us |       0.54% |  11.259 us |       0.49% |  -0.006 us |  -0.05% |  🔵 SAME  |
|   I16   |      I32      |      2^20      |  13.232 us |       2.81% |  13.251 us |       2.38% |   0.019 us |   0.14% |  🔵 SAME  |
|   I16   |      I32      |      2^24      |  28.230 us |       3.70% |  28.385 us |       3.75% |   0.155 us |   0.55% |  🔵 SAME  |
|   I16   |      I32      |      2^28      | 213.127 us |       0.56% | 213.061 us |       0.54% |  -0.066 us |  -0.03% |  🔵 SAME  |
|   I16   |      I64      |      2^16      |  11.261 us |       0.64% |  11.259 us |       0.47% |  -0.002 us |  -0.02% |  🔵 SAME  |
|   I16   |      I64      |      2^20      |  12.986 us |       5.73% |  12.938 us |       6.06% |  -0.048 us |  -0.37% |  🔵 SAME  |
|   I16   |      I64      |      2^24      |  29.972 us |       3.76% |  30.011 us |       3.50% |   0.039 us |   0.13% |  🔵 SAME  |
|   I16   |      I64      |      2^28      | 211.505 us |       0.62% | 211.674 us |       0.61% |   0.169 us |   0.08% |  🔵 SAME  |
|   I16   |      I64      |      2^32      |   3.249 ms |       1.51% |   3.285 ms |       1.47% |  35.835 us |   1.10% |  🔵 SAME  |
|   I32   |      I32      |      2^16      |   9.832 us |       9.53% |   9.656 us |       8.77% |  -0.176 us |  -1.79% |  🔵 SAME  |
|   I32   |      I32      |      2^20      |  13.182 us |       4.04% |  13.152 us |       4.54% |  -0.030 us |  -0.23% |  🔵 SAME  |
|   I32   |      I32      |      2^24      |  35.789 us |       3.24% |  35.818 us |       3.30% |   0.028 us |   0.08% |  🔵 SAME  |
|   I32   |      I32      |      2^28      | 322.907 us |       0.50% | 322.808 us |       0.55% |  -0.099 us |  -0.03% |  🔵 SAME  |
|   I32   |      I64      |      2^16      |   9.723 us |       9.06% |   9.660 us |       8.86% |  -0.063 us |  -0.65% |  🔵 SAME  |
|   I32   |      I64      |      2^20      |  12.948 us |       6.10% |  12.949 us |       6.69% |   0.001 us |   0.01% |  🔵 SAME  |
|   I32   |      I64      |      2^24      |  35.871 us |       3.19% |  35.910 us |       3.18% |   0.039 us |   0.11% |  🔵 SAME  |
|   I32   |      I64      |      2^28      | 322.872 us |       0.55% | 322.683 us |       0.50% |  -0.188 us |  -0.06% |  🔵 SAME  |
|   I32   |      I64      |      2^32      |   4.994 ms |       0.92% |   5.038 ms |       1.15% |  43.860 us |   0.88% |  🔵 SAME  |
|   I64   |      I32      |      2^16      |  11.266 us |       1.19% |  11.266 us |       0.95% |   0.000 us |   0.00% |  🔵 SAME  |
|   I64   |      I32      |      2^20      |  15.943 us |       5.87% |  16.027 us |       6.08% |   0.084 us |   0.53% |  🔵 SAME  |
|   I64   |      I32      |      2^24      |  54.290 us |       1.80% |  54.602 us |       1.81% |   0.311 us |   0.57% |  🔵 SAME  |
|   I64   |      I32      |      2^28      | 625.374 us |       0.29% | 625.037 us |       0.28% |  -0.338 us |  -0.05% |  🔵 SAME  |
|   I64   |      I64      |      2^16      |  11.262 us |       0.51% |  11.257 us |       0.55% |  -0.005 us |  -0.04% |  🔵 SAME  |
|   I64   |      I64      |      2^20      |  15.484 us |       3.38% |  15.489 us |       3.34% |   0.005 us |   0.03% |  🔵 SAME  |
|   I64   |      I64      |      2^24      |  54.337 us |       1.80% |  54.409 us |       1.80% |   0.072 us |   0.13% |  🔵 SAME  |
|   I64   |      I64      |      2^28      | 625.582 us |       0.28% | 625.186 us |       0.27% |  -0.396 us |  -0.06% |  🔵 SAME  |
|   I64   |      I64      |      2^32      |  10.251 ms |       0.95% |  10.338 ms |       0.78% |  87.303 us |   0.85% |  🔴 SLOW  |
|  I128   |      I32      |      2^16      |  12.610 us |       7.71% |  12.168 us |       8.35% |  -0.442 us |  -3.51% |  🔵 SAME  |
|  I128   |      I32      |      2^20      |  21.542 us |       1.51% |  21.545 us |       1.55% |   0.002 us |   0.01% |  🔵 SAME  |
|  I128   |      I32      |      2^24      | 149.502 us |       0.80% | 149.664 us |       0.77% |   0.161 us |   0.11% |  🔵 SAME  |
|  I128   |      I32      |      2^28      |   2.062 ms |       0.15% |   2.072 ms |       0.14% |  10.677 us |   0.52% |  🔴 SLOW  |
|  I128   |      I64      |      2^16      |  13.178 us |       3.82% |  13.176 us |       3.97% |  -0.002 us |  -0.01% |  🔵 SAME  |
|  I128   |      I64      |      2^20      |  23.254 us |       3.59% |  23.460 us |       2.59% |   0.207 us |   0.89% |  🔵 SAME  |
|  I128   |      I64      |      2^24      | 150.372 us |       0.77% | 150.916 us |       0.80% |   0.545 us |   0.36% |  🔵 SAME  |
|  I128   |      I64      |      2^28      |   2.062 ms |       0.15% |   2.073 ms |       0.14% |  10.965 us |   0.53% |  🔴 SLOW  |
|  I128   |      I64      |      2^32      |  32.600 ms |       0.03% |  32.783 ms |       0.03% | 183.795 us |   0.56% |  🔴 SLOW  |
|   F32   |      I32      |      2^16      |  10.777 us |       8.02% |  10.559 us |       9.13% |  -0.218 us |  -2.03% |  🔵 SAME  |
|   F32   |      I32      |      2^20      |  13.365 us |       3.08% |  13.359 us |       3.52% |  -0.006 us |  -0.04% |  🔵 SAME  |
|   F32   |      I32      |      2^24      |  36.520 us |       2.91% |  36.730 us |       3.02% |   0.210 us |   0.58% |  🔵 SAME  |
|   F32   |      I32      |      2^28      | 320.841 us |       0.43% | 320.978 us |       0.42% |   0.137 us |   0.04% |  🔵 SAME  |
|   F32   |      I64      |      2^16      |  10.567 us |       9.24% |  10.574 us |       9.15% |   0.007 us |   0.07% |  🔵 SAME  |
|   F32   |      I64      |      2^20      |  13.487 us |       4.89% |  13.461 us |       5.17% |  -0.026 us |  -0.19% |  🔵 SAME  |
|   F32   |      I64      |      2^24      |  36.735 us |       3.03% |  36.710 us |       3.09% |  -0.024 us |  -0.07% |  🔵 SAME  |
|   F32   |      I64      |      2^28      | 321.054 us |       0.51% | 321.093 us |       0.49% |   0.038 us |   0.01% |  🔵 SAME  |
|   F32   |      I64      |      2^32      |   4.865 ms |       0.67% |   4.860 ms |       0.63% |  -4.601 us |  -0.09% |  🔵 SAME  |
|   F64   |      I32      |      2^16      |  11.275 us |       1.87% |  11.282 us |       2.30% |   0.007 us |   0.06% |  🔵 SAME  |
|   F64   |      I32      |      2^20      |  16.383 us |       6.27% |  16.265 us |       6.30% |  -0.118 us |  -0.72% |  🔵 SAME  |
|   F64   |      I32      |      2^24      |  56.907 us |       1.98% |  56.826 us |       1.92% |  -0.081 us |  -0.14% |  🔵 SAME  |
|   F64   |      I32      |      2^28      | 623.181 us |       0.25% | 623.058 us |       0.25% |  -0.123 us |  -0.02% |  🔵 SAME  |
|   F64   |      I64      |      2^16      |  11.257 us |       0.48% |  11.257 us |       0.57% |  -0.000 us |  -0.00% |  🔵 SAME  |
|   F64   |      I64      |      2^20      |  16.859 us |       5.40% |  16.675 us |       5.85% |  -0.183 us |  -1.09% |  🔵 SAME  |
|   F64   |      I64      |      2^24      |  56.930 us |       1.97% |  56.748 us |       1.97% |  -0.182 us |  -0.32% |  🔵 SAME  |
|   F64   |      I64      |      2^28      | 623.043 us |       0.23% | 623.061 us |       0.25% |   0.018 us |   0.00% |  🔵 SAME  |
|   F64   |      I64      |      2^32      |   9.784 ms |       0.47% |   9.842 ms |       0.58% |  57.384 us |   0.59% |  🔴 SLOW  |
|   C32   |      I32      |      2^16      |  11.305 us |       2.64% |  11.313 us |       3.03% |   0.007 us |   0.07% |  🔵 SAME  |
|   C32   |      I32      |      2^20      |  15.994 us |       6.00% |  15.860 us |       5.63% |  -0.134 us |  -0.84% |  🔵 SAME  |
|   C32   |      I32      |      2^24      |  55.027 us |       2.06% |  54.961 us |       2.03% |  -0.067 us |  -0.12% |  🔵 SAME  |
|   C32   |      I32      |      2^28      | 622.923 us |       0.22% | 622.674 us |       0.23% |  -0.249 us |  -0.04% |  🔵 SAME  |
|   C32   |      I64      |      2^16      |  11.257 us |       0.48% |  11.261 us |       0.49% |   0.004 us |   0.04% |  🔵 SAME  |
|   C32   |      I64      |      2^20      |  15.760 us |       5.28% |  15.676 us |       4.77% |  -0.084 us |  -0.53% |  🔵 SAME  |
|   C32   |      I64      |      2^24      |  54.998 us |       2.13% |  54.849 us |       1.86% |  -0.149 us |  -0.27% |  🔵 SAME  |
|   C32   |      I64      |      2^28      | 623.031 us |       0.24% | 622.876 us |       0.22% |  -0.155 us |  -0.02% |  🔵 SAME  |
|   C32   |      I64      |      2^32      |   9.709 ms |       0.04% |   9.708 ms |       0.03% |  -1.453 us |  -0.01% |  🔵 SAME  |
|   F16   |      I32      |      2^16      |  11.271 us |       1.61% |  11.268 us |       1.55% |  -0.003 us |  -0.02% |  🔵 SAME  |
|   F16   |      I32      |      2^20      |  13.254 us |       2.28% |  13.311 us |       1.94% |   0.056 us |   0.43% |  🔵 SAME  |
|   F16   |      I32      |      2^24      |  29.116 us |       3.65% |  29.015 us |       3.91% |  -0.101 us |  -0.35% |  🔵 SAME  |
|   F16   |      I32      |      2^28      | 225.664 us |       0.53% | 225.652 us |       0.52% |  -0.012 us |  -0.01% |  🔵 SAME  |
|   F16   |      I64      |      2^16      |  11.260 us |       0.53% |  11.255 us |       0.53% |  -0.005 us |  -0.05% |  🔵 SAME  |
|   F16   |      I64      |      2^20      |  13.304 us |       1.74% |  13.320 us |       1.45% |   0.017 us |   0.12% |  🔵 SAME  |
|   F16   |      I64      |      2^24      |  29.262 us |       3.71% |  29.194 us |       3.76% |  -0.067 us |  -0.23% |  🔵 SAME  |
|   F16   |      I64      |      2^28      | 225.751 us |       0.52% | 225.709 us |       0.52% |  -0.042 us |  -0.02% |  🔵 SAME  |
|   F16   |      I64      |      2^32      |   3.338 ms |       0.70% |   3.339 ms |       0.71% |   0.699 us |   0.02% |  🔵 SAME  |
|  BF16   |      I32      |      2^16      |  11.261 us |       0.64% |  11.259 us |       0.55% |  -0.002 us |  -0.02% |  🔵 SAME  |
|  BF16   |      I32      |      2^20      |  13.292 us |       1.05% |  13.305 us |       2.17% |   0.012 us |   0.09% |  🔵 SAME  |
|  BF16   |      I32      |      2^24      |  29.141 us |       3.74% |  29.186 us |       3.87% |   0.045 us |   0.15% |  🔵 SAME  |
|  BF16   |      I32      |      2^28      | 225.623 us |       0.52% | 225.722 us |       0.50% |   0.100 us |   0.04% |  🔵 SAME  |
|  BF16   |      I64      |      2^16      |  11.259 us |       0.69% |  11.258 us |       0.51% |  -0.000 us |  -0.00% |  🔵 SAME  |
|  BF16   |      I64      |      2^20      |  13.269 us |       2.31% |  13.290 us |       1.94% |   0.021 us |   0.16% |  🔵 SAME  |
|  BF16   |      I64      |      2^24      |  29.213 us |       3.62% |  29.043 us |       3.69% |  -0.170 us |  -0.58% |  🔵 SAME  |
|  BF16   |      I64      |      2^28      | 225.783 us |       0.51% | 225.584 us |       0.54% |  -0.199 us |  -0.09% |  🔵 SAME  |
|  BF16   |      I64      |      2^32      |   3.338 ms |       0.80% |   3.338 ms |       0.66% |  -0.135 us |  -0.00% |  🔵 SAME  |

# Summary

- Total Matches: 90
  - Pass    (diff <= min_noise): 85
  - Unknown (infinite noise):    0
  - Failure (diff > min_noise):  5

@bernhardmgruber

Copy link
Copy Markdown
Contributor Author

Closed in favor of: #9858

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

Archived in project

Development

Successfully merging this pull request may close these issues.

scan_exclusive_sum regresses in CCCL 3.4 on Ampere

5 participants