CPU Benchmark Results

Target profile: mac-cpu
tenferro-rs commit: 8a1839febeb3c868a502e26397ca10761bbc568d
Benchmark commit: 5df994b8aee8e926b52e885ab89a5732e4e2ff8f

Generated from the explicit runs below; each section retains its own provenance and thread settings.

Threads: 1

Source report: data/results/mac-cpu/cpu/einsum/20260923_201832/cpu_ops_report.md.

Suite: cpu/cpu_ops
Target profile: mac-cpu
Timestamp: 20260923_201832

Latest run: ./scripts/run_all.sh 1.

This file is generated from one CPU ops run under data/results/mac-cpu/cpu/einsum/20260923_201832.

tenferro-rs commit: 8a1839febeb3c868a502e26397ca10761bbc568d

CPU Information

Model: Apple M5 Max
Vendor: Apple
Logical CPUs: 18
Physical CPUs: 18
Sockets: 1
Cores per socket: 18
Threads per core: 1
NUMA nodes: 1
Performance levels: Super: 6 physical / 6 logical (L1i 192 KiB, L1d 128 KiB, L2 16 MiB, 6 CPUs/L2); Performance: 12 physical / 12 logical (L1i 128 KiB, L1d 64 KiB, L2 8 MiB, 6 CPUs/L2)
Python platform: macOS-26.5.1-arm64-arm-64bit-Mach-O

Thread Environment

OMP_NUM_THREADS: 1
OMP_THREAD_LIMIT: 1
OMP_DYNAMIC: FALSE
RAYON_NUM_THREADS: 1
OPENBLAS_NUM_THREADS: 1
GOTO_NUM_THREADS: 1
MKL_NUM_THREADS: 1
VECLIB_MAXIMUM_THREADS: 1
VECLIB_NUM_THREADS: 1
NUMEXPR_NUM_THREADS: 1
BLIS_NUM_THREADS: 1
XLA_FLAGS: --xla_cpu_multi_thread_eigen=false intra_op_parallelism_threads=1

Tenferro CPU BLAS Backend

tenferro-rs features: system-accelerate
TENFERRO_CPU_BACKEND_KIND: blas
BLAS implementation: accelerate
BLAS version: unknown
BLAS library: /System/Library/Frameworks/Accelerate.framework

Python Backend Providers

PyTorch: BLAS provider accelerate, version 2.12.0, BLAS_INFO accelerate, LAPACK_INFO accelerate
linked BLAS/LAPACK libs: /System/Library/Frameworks/Accelerate.framework/Versions/A/Accelerate; @rpath/libomp.dylib
JAX: dot backend xla_cpu, version 0.10.1, jaxlib 0.10.1, default backend cpu, LAPACK provider internal_lapack
linked LAPACK libs: bazel-out/darwin_arm64-opt/bin/jaxlib/cpu/_lapack.so
CSV: data/results/mac-cpu/cpu/einsum/20260923_201832/cpu_ops_t1_20260923_201832.csv
Source table: data/results/mac-cpu/cpu/einsum/20260923_201832/cpu_ops_t1_20260923_201832.md

CPU Benchmark Items

Small rows use time/memory-bounded batches, normalized per operation. Tenferro eager/trace share a CPU execution scope outside timing. Raw JSONL records contain batch durations and operation counts.

Median ± IQR (ms). Missing backends are shown as -.

suite	benchmark	dtype	threads	shape	tenferro-rs direct API (ms)	tenferro-rs eager mode (ms)	tenferro-rs trace mode (ms)	PyTorch Python (ms)	JAX Python (XLA CPU) (ms)	Julia (Base/LinearAlgebra) (ms)	Julia (Strided.jl) (ms)
batched	`batched_eigh`	f64	1	`16x16xbatch1024 (native batch layout)`	-	9.873 ± 0.053	9.830 ± 0.032	9.833 ± 0.141	10.261 ± 0.091	-	-
batched	`batched_eigh`	f64	1	`16x16xbatch16 (native batch layout)`	-	0.163 ± 0.002	0.160 ± 0.005	0.157 ± 0.004	0.157 ± 0.001	-	-
batched	`batched_eigh`	f64	1	`16x16xbatch256 (native batch layout)`	-	2.465 ± 0.016	2.473 ± 0.034	2.468 ± 0.020	2.532 ± 0.027	-	-
batched	`batched_eigh`	f64	1	`16x16xbatch64 (native batch layout)`	-	0.618 ± 0.003	0.621 ± 0.013	0.621 ± 0.006	0.630 ± 0.006	-	-
batched	`batched_eigh`	f64	1	`2x2xbatch1024 (native batch layout)`	-	0.122 ± 0.001	0.121 ± 0.001	0.129 ± 0.002	0.265 ± 0.008	-	-
batched	`batched_eigh`	f64	1	`2x2xbatch16 (native batch layout)`	-	0.005 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.008 ± 0.000	-	-
batched	`batched_eigh`	f64	1	`2x2xbatch256 (native batch layout)`	-	0.033 ± 0.001	0.034 ± 0.001	0.034 ± 0.001	0.070 ± 0.003	-	-
batched	`batched_eigh`	f64	1	`2x2xbatch64 (native batch layout)`	-	0.011 ± 0.000	0.012 ± 0.000	0.010 ± 0.000	0.019 ± 0.001	-	-
batched	`batched_eigh`	f64	1	`4x4xbatch1024 (native batch layout)`	-	0.665 ± 0.012	0.674 ± 0.016	0.676 ± 0.014	0.809 ± 0.015	-	-
batched	`batched_eigh`	f64	1	`4x4xbatch16 (native batch layout)`	-	0.014 ± 0.000	0.015 ± 0.001	0.013 ± 0.000	0.016 ± 0.000	-	-
batched	`batched_eigh`	f64	1	`4x4xbatch256 (native batch layout)`	-	0.169 ± 0.002	0.171 ± 0.003	0.171 ± 0.003	0.204 ± 0.002	-	-
batched	`batched_eigh`	f64	1	`4x4xbatch64 (native batch layout)`	-	0.045 ± 0.001	0.046 ± 0.001	0.045 ± 0.001	0.055 ± 0.001	-	-
batched	`batched_eigh`	f64	1	`8x8xbatch1024 (native batch layout)`	-	2.570 ± 0.038	2.563 ± 0.056	2.579 ± 0.061	2.722 ± 0.035	-	-
batched	`batched_eigh`	f64	1	`8x8xbatch16 (native batch layout)`	-	0.043 ± 0.001	0.044 ± 0.001	0.042 ± 0.001	0.046 ± 0.001	-	-
batched	`batched_eigh`	f64	1	`8x8xbatch256 (native batch layout)`	-	0.641 ± 0.014	0.640 ± 0.011	0.640 ± 0.015	0.680 ± 0.009	-	-
batched	`batched_eigh`	f64	1	`8x8xbatch64 (native batch layout)`	-	0.162 ± 0.004	0.164 ± 0.004	0.162 ± 0.003	0.170 ± 0.002	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`16x16xbatch1024 (native batch layout)`	-	0.243 ± 0.026	0.302 ± 0.002	1.000 ± 0.053	0.348 ± 0.027	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`16x16xbatch16 (native batch layout)`	-	0.007 ± 0.000	0.011 ± 0.000	0.022 ± 0.001	0.043 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`16x16xbatch256 (native batch layout)`	-	0.068 ± 0.002	0.082 ± 0.000	0.250 ± 0.009	0.073 ± 0.004	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`16x16xbatch64 (native batch layout)`	-	0.019 ± 0.000	0.028 ± 0.001	0.065 ± 0.003	0.049 ± 0.003	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`2x2xbatch1024 (native batch layout)`	-	0.030 ± 0.001	0.033 ± 0.001	0.008 ± 0.000	0.043 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`2x2xbatch16 (native batch layout)`	-	0.003 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	0.037 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`2x2xbatch256 (native batch layout)`	-	0.009 ± 0.000	0.011 ± 0.000	0.005 ± 0.000	0.041 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`2x2xbatch64 (native batch layout)`	-	0.004 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.041 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`4x4xbatch1024 (native batch layout)`	-	0.038 ± 0.002	0.045 ± 0.001	0.018 ± 0.001	0.047 ± 0.003	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`4x4xbatch16 (native batch layout)`	-	0.003 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	0.042 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`4x4xbatch256 (native batch layout)`	-	0.011 ± 0.001	0.015 ± 0.001	0.008 ± 0.000	0.042 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`4x4xbatch64 (native batch layout)`	-	0.004 ± 0.000	0.007 ± 0.000	0.005 ± 0.000	0.041 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`8x8xbatch1024 (native batch layout)`	-	0.085 ± 0.003	0.106 ± 0.004	0.480 ± 0.017	0.068 ± 0.004	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`8x8xbatch16 (native batch layout)`	-	0.003 ± 0.000	0.006 ± 0.000	0.012 ± 0.000	0.041 ± 0.001	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`8x8xbatch256 (native batch layout)`	-	0.023 ± 0.000	0.030 ± 0.000	0.125 ± 0.003	0.047 ± 0.002	-	-
batched	`batched_matmul_ikb_kjb_ijb`	f64	1	`8x8xbatch64 (native batch layout)`	-	0.007 ± 0.000	0.011 ± 0.000	0.034 ± 0.002	0.042 ± 0.001	-	-
batched	`batched_qr`	f64	1	`16x16xbatch1024 (native batch layout)`	-	1.824 ± 0.022	1.835 ± 0.046	1.846 ± 0.023	1.945 ± 0.057	-	-
batched	`batched_qr`	f64	1	`16x16xbatch16 (native batch layout)`	-	0.033 ± 0.001	0.034 ± 0.001	0.030 ± 0.001	0.036 ± 0.001	-	-
batched	`batched_qr`	f64	1	`16x16xbatch256 (native batch layout)`	-	0.462 ± 0.010	0.474 ± 0.005	0.455 ± 0.018	0.486 ± 0.012	-	-
batched	`batched_qr`	f64	1	`16x16xbatch64 (native batch layout)`	-	0.121 ± 0.003	0.120 ± 0.002	0.116 ± 0.002	0.124 ± 0.002	-	-
batched	`batched_qr`	f64	1	`2x2xbatch1024 (native batch layout)`	-	0.124 ± 0.002	0.125 ± 0.002	0.118 ± 0.001	0.158 ± 0.005	-	-
batched	`batched_qr`	f64	1	`2x2xbatch16 (native batch layout)`	-	0.005 ± 0.000	0.006 ± 0.000	0.003 ± 0.000	0.006 ± 0.000	-	-
batched	`batched_qr`	f64	1	`2x2xbatch256 (native batch layout)`	-	0.034 ± 0.000	0.035 ± 0.001	0.031 ± 0.000	0.043 ± 0.001	-	-
batched	`batched_qr`	f64	1	`2x2xbatch64 (native batch layout)`	-	0.011 ± 0.000	0.012 ± 0.000	0.009 ± 0.000	0.013 ± 0.000	-	-
batched	`batched_qr`	f64	1	`4x4xbatch1024 (native batch layout)`	-	0.270 ± 0.003	0.276 ± 0.003	0.272 ± 0.007	0.318 ± 0.003	-	-
batched	`batched_qr`	f64	1	`4x4xbatch16 (native batch layout)`	-	0.008 ± 0.000	0.008 ± 0.000	0.006 ± 0.000	0.009 ± 0.000	-	-
batched	`batched_qr`	f64	1	`4x4xbatch256 (native batch layout)`	-	0.070 ± 0.001	0.072 ± 0.001	0.070 ± 0.001	0.082 ± 0.001	-	-
batched	`batched_qr`	f64	1	`4x4xbatch64 (native batch layout)`	-	0.020 ± 0.000	0.021 ± 0.000	0.019 ± 0.000	0.025 ± 0.000	-	-
batched	`batched_qr`	f64	1	`8x8xbatch1024 (native batch layout)`	-	0.672 ± 0.013	0.676 ± 0.011	0.677 ± 0.024	0.732 ± 0.008	-	-
batched	`batched_qr`	f64	1	`8x8xbatch16 (native batch layout)`	-	0.014 ± 0.000	0.015 ± 0.000	0.012 ± 0.000	0.014 ± 0.000	-	-
batched	`batched_qr`	f64	1	`8x8xbatch256 (native batch layout)`	-	0.174 ± 0.004	0.173 ± 0.004	0.171 ± 0.004	0.184 ± 0.002	-	-
batched	`batched_qr`	f64	1	`8x8xbatch64 (native batch layout)`	-	0.045 ± 0.001	0.047 ± 0.000	0.044 ± 0.001	0.051 ± 0.001	-	-
batched	`batched_solve`	f64	1	`16x16xbatch1024 (native batch layout),rhs=1`	-	0.588 ± 0.017	0.580 ± 0.032	0.602 ± 0.040	0.740 ± 0.032	-	-
batched	`batched_solve`	f64	1	`16x16xbatch16 (native batch layout),rhs=1`	-	0.015 ± 0.001	0.015 ± 0.000	0.014 ± 0.001	0.016 ± 0.000	-	-
batched	`batched_solve`	f64	1	`16x16xbatch256 (native batch layout),rhs=1`	-	0.150 ± 0.006	0.149 ± 0.003	0.153 ± 0.004	0.175 ± 0.005	-	-
batched	`batched_solve`	f64	1	`16x16xbatch64 (native batch layout),rhs=1`	-	0.041 ± 0.002	0.043 ± 0.001	0.042 ± 0.002	0.047 ± 0.000	-	-
batched	`batched_solve`	f64	1	`2x2xbatch1024 (native batch layout),rhs=1`	-	0.055 ± 0.001	0.055 ± 0.002	0.059 ± 0.001	0.075 ± 0.002	-	-
batched	`batched_solve`	f64	1	`2x2xbatch16 (native batch layout),rhs=1`	-	0.005 ± 0.000	0.007 ± 0.000	0.005 ± 0.000	0.005 ± 0.000	-	-
batched	`batched_solve`	f64	1	`2x2xbatch256 (native batch layout),rhs=1`	-	0.017 ± 0.000	0.018 ± 0.000	0.018 ± 0.000	0.023 ± 0.001	-	-
batched	`batched_solve`	f64	1	`2x2xbatch64 (native batch layout),rhs=1`	-	0.008 ± 0.000	0.009 ± 0.000	0.008 ± 0.000	0.007 ± 0.000	-	-
batched	`batched_solve`	f64	1	`4x4xbatch1024 (native batch layout),rhs=1`	-	0.088 ± 0.004	0.094 ± 0.002	0.096 ± 0.004	0.110 ± 0.003	-	-
batched	`batched_solve`	f64	1	`4x4xbatch16 (native batch layout),rhs=1`	-	0.006 ± 0.000	0.007 ± 0.000	0.006 ± 0.000	0.005 ± 0.000	-	-
batched	`batched_solve`	f64	1	`4x4xbatch256 (native batch layout),rhs=1`	-	0.025 ± 0.001	0.028 ± 0.001	0.028 ± 0.001	0.033 ± 0.001	-	-
batched	`batched_solve`	f64	1	`4x4xbatch64 (native batch layout),rhs=1`	-	0.010 ± 0.000	0.011 ± 0.000	0.011 ± 0.000	0.010 ± 0.000	-	-
batched	`batched_solve`	f64	1	`8x8xbatch1024 (native batch layout),rhs=1`	-	0.196 ± 0.009	0.198 ± 0.005	0.212 ± 0.008	0.242 ± 0.010	-	-
batched	`batched_solve`	f64	1	`8x8xbatch16 (native batch layout),rhs=1`	-	0.008 ± 0.000	0.009 ± 0.000	0.008 ± 0.000	0.007 ± 0.000	-	-
batched	`batched_solve`	f64	1	`8x8xbatch256 (native batch layout),rhs=1`	-	0.055 ± 0.001	0.054 ± 0.003	0.057 ± 0.001	0.062 ± 0.003	-	-
batched	`batched_solve`	f64	1	`8x8xbatch64 (native batch layout),rhs=1`	-	0.017 ± 0.000	0.019 ± 0.001	0.018 ± 0.001	0.020 ± 0.000	-	-
batched	`batched_svd`	f64	1	`16x16xbatch1024 (native batch layout)`	-	19.923 ± 0.183	19.916 ± 0.105	20.010 ± 0.116	20.646 ± 0.124	-	-
batched	`batched_svd`	f64	1	`16x16xbatch16 (native batch layout)`	-	0.317 ± 0.005	0.320 ± 0.008	0.318 ± 0.005	0.321 ± 0.005	-	-
batched	`batched_svd`	f64	1	`16x16xbatch256 (native batch layout)`	-	5.001 ± 0.190	5.070 ± 0.074	4.973 ± 0.065	5.102 ± 0.065	-	-
batched	`batched_svd`	f64	1	`16x16xbatch64 (native batch layout)`	-	1.269 ± 0.030	1.250 ± 0.016	1.250 ± 0.028	1.287 ± 0.014	-	-
batched	`batched_svd`	f64	1	`2x2xbatch1024 (native batch layout)`	-	0.308 ± 0.004	0.310 ± 0.004	0.314 ± 0.006	1.036 ± 0.007	-	-
batched	`batched_svd`	f64	1	`2x2xbatch16 (native batch layout)`	-	0.009 ± 0.000	0.010 ± 0.000	0.007 ± 0.000	0.021 ± 0.001	-	-
batched	`batched_svd`	f64	1	`2x2xbatch256 (native batch layout)`	-	0.081 ± 0.002	0.083 ± 0.001	0.079 ± 0.001	0.247 ± 0.008	-	-
batched	`batched_svd`	f64	1	`2x2xbatch64 (native batch layout)`	-	0.024 ± 0.000	0.025 ± 0.000	0.022 ± 0.001	0.069 ± 0.002	-	-
batched	`batched_svd`	f64	1	`4x4xbatch1024 (native batch layout)`	-	1.341 ± 0.020	1.359 ± 0.023	1.373 ± 0.028	1.993 ± 0.049	-	-
batched	`batched_svd`	f64	1	`4x4xbatch16 (native batch layout)`	-	0.025 ± 0.001	0.027 ± 0.000	0.024 ± 0.001	0.036 ± 0.001	-	-
batched	`batched_svd`	f64	1	`4x4xbatch256 (native batch layout)`	-	0.341 ± 0.007	0.346 ± 0.007	0.341 ± 0.007	0.501 ± 0.014	-	-
batched	`batched_svd`	f64	1	`4x4xbatch64 (native batch layout)`	-	0.089 ± 0.002	0.091 ± 0.002	0.087 ± 0.002	0.130 ± 0.002	-	-
batched	`batched_svd`	f64	1	`8x8xbatch1024 (native batch layout)`	-	5.129 ± 0.183	5.100 ± 0.085	5.094 ± 0.098	5.649 ± 0.251	-	-
batched	`batched_svd`	f64	1	`8x8xbatch16 (native batch layout)`	-	0.084 ± 0.002	0.087 ± 0.002	0.083 ± 0.002	0.092 ± 0.001	-	-
batched	`batched_svd`	f64	1	`8x8xbatch256 (native batch layout)`	-	1.270 ± 0.012	1.279 ± 0.023	1.283 ± 0.028	1.411 ± 0.023	-	-
batched	`batched_svd`	f64	1	`8x8xbatch64 (native batch layout)`	-	0.320 ± 0.008	0.332 ± 0.006	0.322 ± 0.011	0.354 ± 0.005	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`16x16xbatch1024 (native batch layout)`	-	0.984 ± 0.063	0.932 ± 0.036	3.860 ± 0.064	0.882 ± 0.037	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`16x16xbatch16 (native batch layout)`	-	0.108 ± 0.001	0.038 ± 0.000	0.095 ± 0.001	0.018 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`16x16xbatch256 (native batch layout)`	-	0.296 ± 0.006	0.255 ± 0.005	0.962 ± 0.020	0.221 ± 0.010	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`16x16xbatch64 (native batch layout)`	-	0.137 ± 0.002	0.092 ± 0.001	0.269 ± 0.004	0.056 ± 0.001	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`2x2xbatch1024 (native batch layout)`	-	0.155 ± 0.003	0.105 ± 0.001	0.039 ± 0.001	0.009 ± 0.001	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`2x2xbatch16 (native batch layout)`	-	0.082 ± 0.002	0.017 ± 0.000	0.028 ± 0.001	0.004 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`2x2xbatch256 (native batch layout)`	-	0.103 ± 0.002	0.038 ± 0.000	0.031 ± 0.001	0.005 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`2x2xbatch64 (native batch layout)`	-	0.085 ± 0.002	0.021 ± 0.000	0.029 ± 0.001	0.005 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`4x4xbatch1024 (native batch layout)`	-	0.199 ± 0.004	0.148 ± 0.003	0.073 ± 0.003	0.174 ± 0.004	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`4x4xbatch16 (native batch layout)`	-	0.080 ± 0.002	0.018 ± 0.000	0.028 ± 0.001	0.005 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`4x4xbatch256 (native batch layout)`	-	0.113 ± 0.001	0.050 ± 0.001	0.040 ± 0.001	0.049 ± 0.001	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`4x4xbatch64 (native batch layout)`	-	0.087 ± 0.002	0.024 ± 0.000	0.031 ± 0.001	0.014 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`8x8xbatch1024 (native batch layout)`	-	0.539 ± 0.007	0.527 ± 0.014	3.189 ± 0.063	0.219 ± 0.008	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`8x8xbatch16 (native batch layout)`	-	0.092 ± 0.001	0.031 ± 0.000	0.083 ± 0.001	0.006 ± 0.000	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`8x8xbatch256 (native batch layout)`	-	0.198 ± 0.002	0.155 ± 0.001	0.827 ± 0.019	0.057 ± 0.001	-	-
batched	`grad_sum_batched_matmul_backward`	f64	1	`8x8xbatch64 (native batch layout)`	-	0.116 ± 0.001	0.062 ± 0.001	0.232 ± 0.003	0.019 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`16x16xbatch1024 (native batch layout),rhs=1`	-	1.041 ± 0.049	1.025 ± 0.041	0.904 ± 0.081	0.876 ± 0.026	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`16x16xbatch16 (native batch layout),rhs=1`	-	0.117 ± 0.001	0.038 ± 0.001	0.035 ± 0.002	0.020 ± 0.001	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`16x16xbatch256 (native batch layout),rhs=1`	-	0.325 ± 0.014	0.276 ± 0.009	0.235 ± 0.010	0.208 ± 0.009	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`16x16xbatch64 (native batch layout),rhs=1`	-	0.152 ± 0.003	0.086 ± 0.002	0.073 ± 0.003	0.056 ± 0.001	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`2x2xbatch1024 (native batch layout),rhs=1`	-	0.195 ± 0.005	0.153 ± 0.005	0.103 ± 0.003	0.101 ± 0.002	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`2x2xbatch16 (native batch layout),rhs=1`	-	0.104 ± 0.006	0.023 ± 0.001	0.022 ± 0.001	0.007 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`2x2xbatch256 (native batch layout),rhs=1`	-	0.128 ± 0.002	0.054 ± 0.001	0.042 ± 0.001	0.031 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`2x2xbatch64 (native batch layout),rhs=1`	-	0.107 ± 0.001	0.029 ± 0.000	0.026 ± 0.000	0.009 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`4x4xbatch1024 (native batch layout),rhs=1`	-	0.253 ± 0.003	0.210 ± 0.003	0.156 ± 0.005	0.149 ± 0.002	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`4x4xbatch16 (native batch layout),rhs=1`	-	0.101 ± 0.002	0.024 ± 0.001	0.023 ± 0.001	0.006 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`4x4xbatch256 (native batch layout),rhs=1`	-	0.143 ± 0.002	0.069 ± 0.000	0.055 ± 0.001	0.043 ± 0.001	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`4x4xbatch64 (native batch layout),rhs=1`	-	0.111 ± 0.002	0.033 ± 0.000	0.029 ± 0.001	0.013 ± 0.001	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`8x8xbatch1024 (native batch layout),rhs=1`	-	0.413 ± 0.014	0.389 ± 0.008	0.313 ± 0.016	0.296 ± 0.011	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`8x8xbatch16 (native batch layout),rhs=1`	-	0.106 ± 0.001	0.027 ± 0.000	0.025 ± 0.000	0.008 ± 0.000	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`8x8xbatch256 (native batch layout),rhs=1`	-	0.175 ± 0.003	0.114 ± 0.003	0.095 ± 0.004	0.077 ± 0.002	-	-
batched	`grad_sum_batched_solve_backward`	f64	1	`8x8xbatch64 (native batch layout),rhs=1`	-	0.123 ± 0.002	0.045 ± 0.001	0.039 ± 0.001	0.026 ± 0.001	-	-
large	`eigh`	f64	1	`1024x1024`	-	63.308 ± 0.267	63.118 ± 0.092	64.132 ± 0.451	67.439 ± 0.412	-	-
large	`eigh`	f64	1	`128x128`	-	0.629 ± 0.004	0.628 ± 0.007	0.625 ± 0.006	0.653 ± 0.006	-	-
large	`eigh`	f64	1	`256x256`	-	2.679 ± 0.014	2.667 ± 0.025	2.703 ± 0.037	2.796 ± 0.034	-	-
large	`eigh`	f64	1	`512x512`	-	12.062 ± 0.093	11.992 ± 0.077	12.257 ± 0.067	12.761 ± 0.066	-	-
large	`eigh`	f64	1	`64x64`	-	0.201 ± 0.001	0.204 ± 0.001	0.201 ± 0.002	0.206 ± 0.002	-	-
large	`grad_sum_eigh_jvp`	f64	1	`1024x1024`	-	-	71.144 ± 0.291	80.812 ± 0.601	134.101 ± 0.255	-	-
large	`grad_sum_eigh_jvp`	f64	1	`256x256`	-	-	2.922 ± 0.031	3.051 ± 0.019	3.897 ± 0.015	-	-
large	`grad_sum_eigh_jvp`	f64	1	`512x512`	-	-	13.298 ± 0.055	14.471 ± 0.098	21.305 ± 0.062	-	-
large	`grad_sum_eigh_vjp`	f64	1	`1024x1024`	-	-	70.883 ± 0.762	69.317 ± 0.848	132.853 ± 0.189	-	-
large	`grad_sum_eigh_vjp`	f64	1	`256x256`	-	-	2.876 ± 0.008	2.798 ± 0.023	3.884 ± 0.018	-	-
large	`grad_sum_eigh_vjp`	f64	1	`512x512`	-	-	13.193 ± 0.052	12.916 ± 0.047	21.123 ± 0.126	-	-
large	`grad_sum_lu_jvp`	f64	1	`1024x1024`	-	-	32.001 ± 1.278	59.123 ± 0.989	85.911 ± 0.266	-	-
large	`grad_sum_lu_jvp`	f64	1	`256x256`	-	-	0.656 ± 0.003	1.295 ± 0.020	1.713 ± 0.005	-	-
large	`grad_sum_lu_jvp`	f64	1	`512x512`	-	-	3.983 ± 0.028	8.359 ± 0.037	11.859 ± 0.067	-	-
large	`grad_sum_lu_vjp`	f64	1	`1024x1024`	-	-	28.797 ± 0.903	30.872 ± 0.674	86.545 ± 0.784	-	-
large	`grad_sum_lu_vjp`	f64	1	`256x256`	-	-	0.678 ± 0.005	0.763 ± 0.007	1.722 ± 0.016	-	-
large	`grad_sum_lu_vjp`	f64	1	`512x512`	-	-	4.002 ± 0.075	4.585 ± 0.067	11.712 ± 0.045	-	-
large	`grad_sum_matmul`	f64	1	`1024x1024`	-	5.201 ± 1.010	5.151 ± 1.242	5.600 ± 0.770	33.293 ± 0.151	-	-
large	`grad_sum_matmul`	f64	1	`128x128`	-	0.031 ± 0.001	0.027 ± 0.000	0.018 ± 0.001	0.077 ± 0.001	-	-
large	`grad_sum_matmul`	f64	1	`256x256`	-	0.117 ± 0.004	0.104 ± 0.003	0.103 ± 0.003	0.562 ± 0.006	-	-
large	`grad_sum_matmul`	f64	1	`512x512`	-	0.706 ± 0.004	0.657 ± 0.026	0.631 ± 0.012	4.232 ± 0.021	-	-
large	`grad_sum_matmul`	f64	1	`64x64`	-	0.012 ± 0.000	0.011 ± 0.000	0.005 ± 0.000	0.014 ± 0.000	-	-
large	`grad_sum_matmul_backward`	f64	1	`1024x1024`	-	16.577 ± 0.775	16.455 ± 0.452	16.323 ± 0.496	100.528 ± 0.537	-	-
large	`grad_sum_matmul_backward`	f64	1	`128x128`	-	0.126 ± 0.002	0.081 ± 0.001	0.059 ± 0.001	0.223 ± 0.001	-	-
large	`grad_sum_matmul_backward`	f64	1	`256x256`	-	0.397 ± 0.004	0.333 ± 0.003	0.294 ± 0.010	1.666 ± 0.021	-	-
large	`grad_sum_matmul_backward`	f64	1	`512x512`	-	2.250 ± 0.057	2.127 ± 0.020	1.938 ± 0.063	12.647 ± 0.053	-	-
large	`grad_sum_matmul_backward`	f64	1	`64x64`	-	0.101 ± 0.001	0.033 ± 0.000	0.024 ± 0.001	0.036 ± 0.000	-	-
large	`grad_sum_qr_jvp`	f64	1	`1024x1024`	-	-	49.817 ± 0.552	56.094 ± 0.538	134.247 ± 0.251	-	-
large	`grad_sum_qr_jvp`	f64	1	`256x256`	-	-	1.787 ± 0.010	1.909 ± 0.039	3.145 ± 0.021	-	-
large	`grad_sum_qr_jvp`	f64	1	`512x512`	-	-	8.538 ± 0.037	9.843 ± 0.055	18.737 ± 0.059	-	-
large	`grad_sum_qr_vjp`	f64	1	`1024x1024`	-	-	50.146 ± 0.598	53.597 ± 1.955	136.868 ± 3.204	-	-
large	`grad_sum_qr_vjp`	f64	1	`256x256`	-	-	1.829 ± 0.009	1.834 ± 0.011	3.122 ± 0.012	-	-
large	`grad_sum_qr_vjp`	f64	1	`512x512`	-	-	8.708 ± 0.019	9.324 ± 0.068	19.406 ± 0.083	-	-
large	`grad_sum_solve_backward`	f64	1	`1024x1024,rhs=1`	-	5.412 ± 0.121	5.070 ± 0.103	5.812 ± 0.117	5.103 ± 0.065	-	-
large	`grad_sum_solve_backward`	f64	1	`128x128,rhs=1`	-	0.143 ± 0.002	0.078 ± 0.001	0.077 ± 0.003	0.061 ± 0.002	-	-
large	`grad_sum_solve_backward`	f64	1	`256x256,rhs=1`	-	0.304 ± 0.010	0.242 ± 0.003	0.256 ± 0.011	0.237 ± 0.011	-	-
large	`grad_sum_solve_backward`	f64	1	`512x512,rhs=1`	-	1.177 ± 0.042	1.063 ± 0.017	1.235 ± 0.032	1.025 ± 0.028	-	-
large	`grad_sum_solve_backward`	f64	1	`64x64,rhs=1`	-	0.123 ± 0.002	0.040 ± 0.000	0.039 ± 0.001	0.021 ± 0.001	-	-
large	`grad_sum_solve_jvp`	f64	1	`1024x1024,rhs=1`	-	-	4.753 ± 0.050	5.611 ± 0.085	5.165 ± 0.054	-	-
large	`grad_sum_solve_jvp`	f64	1	`256x256,rhs=1`	-	-	0.216 ± 0.011	0.293 ± 0.008	0.220 ± 0.002	-	-
large	`grad_sum_solve_jvp`	f64	1	`512x512,rhs=1`	-	-	1.000 ± 0.045	1.182 ± 0.046	1.019 ± 0.016	-	-
large	`grad_sum_solve_vjp`	f64	1	`1024x1024,rhs=1`	-	-	4.629 ± 0.074	9.263 ± 0.041	4.752 ± 0.040	-	-
large	`grad_sum_solve_vjp`	f64	1	`256x256,rhs=1`	-	-	0.209 ± 0.001	0.387 ± 0.006	0.215 ± 0.002	-	-
large	`grad_sum_solve_vjp`	f64	1	`512x512,rhs=1`	-	-	0.942 ± 0.016	1.862 ± 0.010	0.978 ± 0.013	-	-
large	`grad_sum_svd_s_backward`	f64	1	`1024x1024`	-	116.525 ± 1.083	115.488 ± 0.629	112.287 ± 1.812	175.515 ± 1.598	-	-
large	`grad_sum_svd_s_backward`	f64	1	`128x128`	-	1.304 ± 0.014	1.244 ± 0.008	1.254 ± 0.014	1.368 ± 0.012	-	-
large	`grad_sum_svd_s_backward`	f64	1	`256x256`	-	5.095 ± 0.045	5.010 ± 0.026	4.946 ± 0.038	6.009 ± 0.019	-	-
large	`grad_sum_svd_s_backward`	f64	1	`512x512`	-	23.029 ± 0.084	22.906 ± 0.052	22.530 ± 0.104	30.639 ± 0.134	-	-
large	`grad_sum_svd_s_backward`	f64	1	`64x64`	-	0.351 ± 0.007	0.300 ± 0.004	0.298 ± 0.003	0.300 ± 0.001	-	-
large	`grad_sum_svd_s_jvp`	f64	1	`1024x1024`	-	-	116.999 ± 0.950	133.649 ± 1.495	176.614 ± 0.626	-	-
large	`grad_sum_svd_s_jvp`	f64	1	`256x256`	-	-	5.012 ± 0.036	5.563 ± 0.094	6.051 ± 0.024	-	-
large	`grad_sum_svd_s_jvp`	f64	1	`512x512`	-	-	22.796 ± 0.103	25.756 ± 0.142	30.771 ± 0.044	-	-
large	`grad_sum_svd_s_vjp`	f64	1	`1024x1024`	-	-	115.746 ± 0.731	112.015 ± 1.212	175.465 ± 0.389	-	-
large	`grad_sum_svd_s_vjp`	f64	1	`256x256`	-	-	5.009 ± 0.029	5.054 ± 0.063	6.009 ± 0.024	-	-
large	`grad_sum_svd_s_vjp`	f64	1	`512x512`	-	-	22.843 ± 0.096	22.482 ± 0.149	30.517 ± 0.051	-	-
large	`matmul`	f64	1	`1024x1024`	-	4.523 ± 0.068	5.152 ± 0.696	4.931 ± 0.296	32.884 ± 0.151	-	-
large	`matmul`	f64	1	`128x128`	-	0.014 ± 0.000	0.022 ± 0.000	0.014 ± 0.001	0.069 ± 0.001	-	-
large	`matmul`	f64	1	`256x256`	-	0.085 ± 0.001	0.101 ± 0.002	0.097 ± 0.002	0.537 ± 0.004	-	-
large	`matmul`	f64	1	`512x512`	-	0.565 ± 0.009	0.658 ± 0.012	0.602 ± 0.018	4.120 ± 0.039	-	-
large	`matmul_rect`	f64	1	`1024x256 * 256x1024`	-	1.210 ± 0.012	1.284 ± 0.024	1.252 ± 0.031	8.075 ± 0.072	-	-
large	`matmul_rect`	f64	1	`256x1024 * 1024x256`	-	0.288 ± 0.005	0.369 ± 0.008	0.345 ± 0.017	2.145 ± 0.027	-	-
large	`qr`	f64	1	`1024x1024`	-	25.416 ± 0.105	25.246 ± 0.063	31.303 ± 0.428	27.152 ± 0.124	-	-
large	`qr`	f64	1	`128x128`	-	0.380 ± 0.002	0.379 ± 0.003	0.401 ± 0.004	0.384 ± 0.004	-	-
large	`qr`	f64	1	`256x256`	-	1.182 ± 0.067	1.181 ± 0.004	1.335 ± 0.016	1.237 ± 0.005	-	-
large	`qr`	f64	1	`512x512`	-	5.013 ± 0.026	5.009 ± 0.022	6.142 ± 0.068	5.381 ± 0.029	-	-
large	`qr`	f64	1	`64x64`	-	0.071 ± 0.000	0.072 ± 0.000	0.070 ± 0.001	0.072 ± 0.001	-	-
large	`solve`	f64	1	`1024x1024,rhs=1`	-	4.083 ± 0.078	4.098 ± 0.030	5.195 ± 0.085	5.009 ± 0.029	-	-
large	`solve`	f64	1	`1024x1024,rhs=16`	-	4.018 ± 0.017	3.990 ± 0.061	5.117 ± 0.099	5.021 ± 0.023	-	-
large	`solve`	f64	1	`1024x1024,rhs=64`	-	4.474 ± 0.133	4.377 ± 0.065	5.519 ± 0.112	5.440 ± 0.060	-	-
large	`solve`	f64	1	`128x128,rhs=1`	-	0.039 ± 0.001	0.049 ± 0.001	0.049 ± 0.001	0.054 ± 0.002	-	-
large	`solve`	f64	1	`128x128,rhs=16`	-	0.042 ± 0.000	0.045 ± 0.002	0.050 ± 0.004	0.082 ± 0.008	-	-
large	`solve`	f64	1	`128x128,rhs=64`	-	0.063 ± 0.000	0.057 ± 0.000	0.062 ± 0.003	0.095 ± 0.007	-	-
large	`solve`	f64	1	`256x256,rhs=1`	-	0.155 ± 0.002	0.164 ± 0.001	0.196 ± 0.005	0.208 ± 0.005	-	-
large	`solve`	f64	1	`256x256,rhs=16`	-	0.157 ± 0.011	0.156 ± 0.003	0.191 ± 0.010	0.221 ± 0.012	-	-
large	`solve`	f64	1	`256x256,rhs=64`	-	0.202 ± 0.007	0.185 ± 0.008	0.224 ± 0.004	0.267 ± 0.012	-	-
large	`solve`	f64	1	`512x512,rhs=1`	-	0.806 ± 0.016	0.814 ± 0.015	1.039 ± 0.022	1.031 ± 0.051	-	-
large	`solve`	f64	1	`512x512,rhs=16`	-	0.798 ± 0.020	0.792 ± 0.022	1.042 ± 0.022	1.035 ± 0.025	-	-
large	`solve`	f64	1	`512x512,rhs=64`	-	0.931 ± 0.014	0.883 ± 0.012	1.160 ± 0.028	1.154 ± 0.039	-	-
large	`solve`	f64	1	`64x64,rhs=1`	-	0.014 ± 0.000	0.018 ± 0.000	0.017 ± 0.001	0.017 ± 0.000	-	-
large	`solve`	f64	1	`64x64,rhs=16`	-	0.017 ± 0.000	0.020 ± 0.001	0.019 ± 0.001	0.023 ± 0.001	-	-
large	`solve`	f64	1	`64x64,rhs=64`	-	0.025 ± 0.000	0.025 ± 0.000	0.025 ± 0.002	0.049 ± 0.008	-	-
large	`svd`	f64	1	`1024x1024`	-	106.646 ± 0.321	106.267 ± 0.322	107.257 ± 0.346	107.474 ± 0.575	-	-
large	`svd`	f64	1	`128x128`	-	1.217 ± 0.005	1.215 ± 0.009	1.208 ± 0.005	1.231 ± 0.013	-	-
large	`svd`	f64	1	`256x256`	-	4.848 ± 0.017	4.832 ± 0.023	4.833 ± 0.045	4.862 ± 0.043	-	-
large	`svd`	f64	1	`512x512`	-	21.638 ± 0.050	21.569 ± 0.124	21.708 ± 0.101	21.992 ± 0.078	-	-
large	`svd`	f64	1	`64x64`	-	0.278 ± 0.002	0.278 ± 0.001	0.276 ± 0.001	0.277 ± 0.002	-	-
small	`eigh`	f64	1	`16x16`	-	0.013 ± 0.000	0.014 ± 0.000	0.011 ± 0.000	0.013 ± 0.000	-	-
small	`eigh`	f64	1	`2x2`	-	0.003 ± 0.000	0.004 ± 0.000	0.002 ± 0.000	0.003 ± 0.000	-	-
small	`eigh`	f64	1	`32x32`	-	0.042 ± 0.000	0.040 ± 0.003	0.040 ± 0.001	0.040 ± 0.001	-	-
small	`eigh`	f64	1	`4x4`	-	0.004 ± 0.000	0.005 ± 0.000	0.002 ± 0.000	0.004 ± 0.000	-	-
small	`eigh`	f64	1	`8x8`	-	0.006 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.006 ± 0.000	-	-
small	`einsum_ij_jk_ik`	f64	1	`16x16`	-	0.003 ± 0.000	0.006 ± 0.000	0.006 ± 0.000	0.041 ± 0.001	-	-
small	`einsum_ij_jk_ik`	f64	1	`2x2`	-	0.002 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	0.036 ± 0.000	-	-
small	`einsum_ij_jk_ik`	f64	1	`32x32`	-	0.003 ± 0.000	0.007 ± 0.000	0.007 ± 0.000	0.042 ± 0.001	-	-
small	`einsum_ij_jk_ik`	f64	1	`4x4`	-	0.002 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	0.036 ± 0.001	-	-
small	`einsum_ij_jk_ik`	f64	1	`8x8`	-	0.002 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	0.041 ± 0.001	-	-
small	`grad_sum_eigh_jvp`	f64	1	`16x16`	-	-	0.040 ± 0.000	0.050 ± 0.001	0.013 ± 0.000	-	-
small	`grad_sum_eigh_jvp`	f64	1	`2x2`	-	-	0.022 ± 0.001	0.033 ± 0.001	0.003 ± 0.000	-	-
small	`grad_sum_eigh_jvp`	f64	1	`32x32`	-	-	0.064 ± 0.001	0.074 ± 0.001	0.045 ± 0.003	-	-
small	`grad_sum_eigh_jvp`	f64	1	`4x4`	-	-	0.023 ± 0.000	0.033 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_eigh_jvp`	f64	1	`8x8`	-	-	0.032 ± 0.000	0.043 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_eigh_vjp`	f64	1	`16x16`	-	-	0.039 ± 0.000	0.045 ± 0.001	0.013 ± 0.000	-	-
small	`grad_sum_eigh_vjp`	f64	1	`2x2`	-	-	0.022 ± 0.000	0.028 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_eigh_vjp`	f64	1	`32x32`	-	-	0.063 ± 0.000	0.068 ± 0.001	0.045 ± 0.003	-	-
small	`grad_sum_eigh_vjp`	f64	1	`4x4`	-	-	0.023 ± 0.000	0.029 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_eigh_vjp`	f64	1	`8x8`	-	-	0.025 ± 0.001	0.037 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_lu_jvp`	f64	1	`16x16`	-	-	0.033 ± 0.000	0.061 ± 0.002	0.005 ± 0.000	-	-
small	`grad_sum_lu_jvp`	f64	1	`2x2`	-	-	0.024 ± 0.000	0.050 ± 0.002	0.004 ± 0.000	-	-
small	`grad_sum_lu_jvp`	f64	1	`32x32`	-	-	0.038 ± 0.000	0.075 ± 0.001	0.015 ± 0.000	-	-
small	`grad_sum_lu_jvp`	f64	1	`4x4`	-	-	0.024 ± 0.000	0.050 ± 0.002	0.004 ± 0.000	-	-
small	`grad_sum_lu_jvp`	f64	1	`8x8`	-	-	0.024 ± 0.000	0.057 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_lu_vjp`	f64	1	`16x16`	-	-	0.032 ± 0.000	0.048 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_lu_vjp`	f64	1	`2x2`	-	-	0.024 ± 0.000	0.039 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_lu_vjp`	f64	1	`32x32`	-	-	0.038 ± 0.000	0.056 ± 0.001	0.015 ± 0.000	-	-
small	`grad_sum_lu_vjp`	f64	1	`4x4`	-	-	0.023 ± 0.001	0.039 ± 0.002	0.004 ± 0.000	-	-
small	`grad_sum_lu_vjp`	f64	1	`8x8`	-	-	0.031 ± 0.000	0.047 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_matmul_backward`	f64	1	`16x16`	-	0.090 ± 0.001	0.021 ± 0.000	0.016 ± 0.000	0.005 ± 0.000	-	-
small	`grad_sum_matmul_backward`	f64	1	`2x2`	-	0.101 ± 0.001	0.015 ± 0.000	0.011 ± 0.000	0.004 ± 0.000	-	-
small	`grad_sum_matmul_backward`	f64	1	`32x32`	-	0.091 ± 0.001	0.022 ± 0.000	0.017 ± 0.000	0.007 ± 0.000	-	-
small	`grad_sum_matmul_backward`	f64	1	`4x4`	-	0.102 ± 0.001	0.015 ± 0.000	0.011 ± 0.000	0.004 ± 0.000	-	-
small	`grad_sum_matmul_backward`	f64	1	`8x8`	-	0.082 ± 0.001	0.019 ± 0.000	0.015 ± 0.000	0.005 ± 0.000	-	-
small	`grad_sum_qr_jvp`	f64	1	`16x16`	-	-	0.050 ± 0.000	0.048 ± 0.001	0.008 ± 0.001	-	-
small	`grad_sum_qr_jvp`	f64	1	`2x2`	-	-	0.039 ± 0.001	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_qr_jvp`	f64	1	`32x32`	-	-	0.064 ± 0.000	0.054 ± 0.001	0.020 ± 0.000	-	-
small	`grad_sum_qr_jvp`	f64	1	`4x4`	-	-	0.039 ± 0.001	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_qr_jvp`	f64	1	`8x8`	-	-	0.040 ± 0.001	0.046 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_qr_vjp`	f64	1	`16x16`	-	-	0.050 ± 0.000	0.047 ± 0.001	0.007 ± 0.000	-	-
small	`grad_sum_qr_vjp`	f64	1	`2x2`	-	-	0.040 ± 0.001	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_qr_vjp`	f64	1	`32x32`	-	-	0.065 ± 0.001	0.054 ± 0.001	0.019 ± 0.000	-	-
small	`grad_sum_qr_vjp`	f64	1	`4x4`	-	-	0.040 ± 0.001	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_qr_vjp`	f64	1	`8x8`	-	-	0.047 ± 0.000	0.045 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_solve_backward`	f64	1	`16x16,rhs=1`	-	0.099 ± 0.001	0.021 ± 0.001	0.019 ± 0.001	0.006 ± 0.000	-	-
small	`grad_sum_solve_backward`	f64	1	`2x2,rhs=1`	-	0.096 ± 0.003	0.020 ± 0.000	0.018 ± 0.000	0.005 ± 0.000	-	-
small	`grad_sum_solve_backward`	f64	1	`32x32,rhs=1`	-	0.113 ± 0.001	0.030 ± 0.000	0.027 ± 0.001	0.008 ± 0.000	-	-
small	`grad_sum_solve_backward`	f64	1	`4x4,rhs=1`	-	0.097 ± 0.002	0.020 ± 0.001	0.019 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_solve_backward`	f64	1	`8x8,rhs=1`	-	0.097 ± 0.003	0.020 ± 0.000	0.019 ± 0.000	0.005 ± 0.000	-	-
small	`grad_sum_solve_jvp`	f64	1	`16x16,rhs=1`	-	-	0.017 ± 0.000	0.082 ± 0.004	0.005 ± 0.000	-	-
small	`grad_sum_solve_jvp`	f64	1	`2x2,rhs=1`	-	-	0.016 ± 0.000	0.081 ± 0.005	0.004 ± 0.000	-	-
small	`grad_sum_solve_jvp`	f64	1	`32x32,rhs=1`	-	-	0.024 ± 0.000	0.090 ± 0.002	0.007 ± 0.000	-	-
small	`grad_sum_solve_jvp`	f64	1	`4x4,rhs=1`	-	-	0.016 ± 0.000	0.081 ± 0.003	0.004 ± 0.000	-	-
small	`grad_sum_solve_jvp`	f64	1	`8x8,rhs=1`	-	-	0.016 ± 0.000	0.081 ± 0.003	0.004 ± 0.000	-	-
small	`grad_sum_solve_vjp`	f64	1	`16x16,rhs=1`	-	-	0.016 ± 0.000	0.038 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_solve_vjp`	f64	1	`2x2,rhs=1`	-	-	0.015 ± 0.000	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_solve_vjp`	f64	1	`32x32,rhs=1`	-	-	0.023 ± 0.000	0.049 ± 0.001	0.007 ± 0.000	-	-
small	`grad_sum_solve_vjp`	f64	1	`4x4,rhs=1`	-	-	0.015 ± 0.000	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_solve_vjp`	f64	1	`8x8,rhs=1`	-	-	0.015 ± 0.000	0.037 ± 0.001	0.004 ± 0.000	-	-
small	`grad_sum_svd_s_backward`	f64	1	`16x16`	-	0.109 ± 0.001	0.039 ± 0.001	0.040 ± 0.001	0.028 ± 0.000	-	-
small	`grad_sum_svd_s_backward`	f64	1	`2x2`	-	0.082 ± 0.002	0.013 ± 0.000	0.013 ± 0.000	0.005 ± 0.000	-	-
small	`grad_sum_svd_s_backward`	f64	1	`32x32`	-	0.156 ± 0.000	0.079 ± 0.001	0.082 ± 0.003	0.076 ± 0.001	-	-
small	`grad_sum_svd_s_backward`	f64	1	`4x4`	-	0.085 ± 0.002	0.014 ± 0.000	0.015 ± 0.000	0.006 ± 0.000	-	-
small	`grad_sum_svd_s_backward`	f64	1	`8x8`	-	0.088 ± 0.002	0.018 ± 0.000	0.019 ± 0.000	0.009 ± 0.000	-	-
small	`grad_sum_svd_s_jvp`	f64	1	`16x16`	-	-	0.040 ± 0.000	0.075 ± 0.001	0.027 ± 0.001	-	-
small	`grad_sum_svd_s_jvp`	f64	1	`2x2`	-	-	0.014 ± 0.000	0.047 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_svd_s_jvp`	f64	1	`32x32`	-	-	0.079 ± 0.001	0.122 ± 0.001	0.074 ± 0.002	-	-
small	`grad_sum_svd_s_jvp`	f64	1	`4x4`	-	-	0.015 ± 0.000	0.048 ± 0.001	0.006 ± 0.000	-	-
small	`grad_sum_svd_s_jvp`	f64	1	`8x8`	-	-	0.022 ± 0.000	0.059 ± 0.001	0.009 ± 0.000	-	-
small	`grad_sum_svd_s_vjp`	f64	1	`16x16`	-	-	0.039 ± 0.000	0.056 ± 0.001	0.026 ± 0.004	-	-
small	`grad_sum_svd_s_vjp`	f64	1	`2x2`	-	-	0.013 ± 0.000	0.030 ± 0.001	0.005 ± 0.000	-	-
small	`grad_sum_svd_s_vjp`	f64	1	`32x32`	-	-	0.078 ± 0.002	0.102 ± 0.001	0.075 ± 0.002	-	-
small	`grad_sum_svd_s_vjp`	f64	1	`4x4`	-	-	0.014 ± 0.000	0.031 ± 0.001	0.006 ± 0.000	-	-
small	`grad_sum_svd_s_vjp`	f64	1	`8x8`	-	-	0.018 ± 0.001	0.035 ± 0.002	0.009 ± 0.000	-	-
small	`matmul`	f64	1	`16x16`	-	0.003 ± 0.000	0.006 ± 0.000	0.001 ± 0.000	0.003 ± 0.000	-	-
small	`matmul`	f64	1	`2x2`	-	0.002 ± 0.000	0.004 ± 0.000	0.000 ± 0.000	0.002 ± 0.000	-	-
small	`matmul`	f64	1	`32x32`	-	0.003 ± 0.000	0.006 ± 0.000	0.001 ± 0.000	0.004 ± 0.000	-	-
small	`matmul`	f64	1	`4x4`	-	0.002 ± 0.000	0.004 ± 0.000	0.000 ± 0.000	0.003 ± 0.000	-	-
small	`matmul`	f64	1	`8x8`	-	0.002 ± 0.000	0.004 ± 0.000	0.001 ± 0.000	0.004 ± 0.000	-	-
small	`qr`	f64	1	`16x16`	-	0.005 ± 0.000	0.006 ± 0.000	0.003 ± 0.000	0.006 ± 0.000	-	-
small	`qr`	f64	1	`2x2`	-	0.003 ± 0.000	0.004 ± 0.000	0.001 ± 0.000	0.003 ± 0.000	-	-
small	`qr`	f64	1	`32x32`	-	0.010 ± 0.000	0.011 ± 0.000	0.008 ± 0.000	0.009 ± 0.000	-	-
small	`qr`	f64	1	`4x4`	-	0.004 ± 0.000	0.004 ± 0.000	0.002 ± 0.000	0.004 ± 0.000	-	-
small	`qr`	f64	1	`8x8`	-	0.004 ± 0.000	0.005 ± 0.000	0.002 ± 0.000	0.004 ± 0.000	-	-
small	`solve`	f64	1	`16x16,rhs=1`	-	0.004 ± 0.000	0.006 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	-	-
small	`solve`	f64	1	`16x16,rhs=4`	-	0.004 ± 0.000	0.006 ± 0.000	0.005 ± 0.000	0.004 ± 0.000	-	-
small	`solve`	f64	1	`2x2,rhs=1`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`solve`	f64	1	`2x2,rhs=4`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`solve`	f64	1	`32x32,rhs=1`	-	0.007 ± 0.000	0.010 ± 0.000	0.008 ± 0.000	0.006 ± 0.000	-	-
small	`solve`	f64	1	`32x32,rhs=4`	-	0.007 ± 0.000	0.010 ± 0.000	0.008 ± 0.000	0.006 ± 0.000	-	-
small	`solve`	f64	1	`4x4,rhs=1`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`solve`	f64	1	`4x4,rhs=4`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`solve`	f64	1	`8x8,rhs=1`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`solve`	f64	1	`8x8,rhs=4`	-	0.003 ± 0.000	0.006 ± 0.000	0.004 ± 0.000	0.003 ± 0.000	-	-
small	`svd`	f64	1	`16x16`	-	0.024 ± 0.000	0.024 ± 0.000	0.022 ± 0.000	0.025 ± 0.000	-	-
small	`svd`	f64	1	`2x2`	-	0.004 ± 0.000	0.005 ± 0.000	0.002 ± 0.000	0.005 ± 0.000	-	-
small	`svd`	f64	1	`32x32`	-	0.072 ± 0.000	0.073 ± 0.000	0.069 ± 0.001	0.071 ± 0.001	-	-
small	`svd`	f64	1	`4x4`	-	0.006 ± 0.000	0.007 ± 0.000	0.004 ± 0.000	0.006 ± 0.000	-	-
small	`svd`	f64	1	`8x8`	-	0.009 ± 0.000	0.010 ± 0.000	0.007 ± 0.000	0.010 ± 0.000	-	-

Cross-Backend Spread Audit

Rows with a successful cell more than 10x faster than the slowest successful cell are flagged for operation-specific review. This cross-backend spread check is a warning, not a correctness verdict.

batched/batched_matmul_ikb_kjb_ijb (f64, threads=1, shape=2x2xbatch16 (native batch layout)): tenferro-eager is 14.7x faster than the slowest successful cell (jax-cpu, 0.037 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/batched_matmul_ikb_kjb_ijb (f64, threads=1, shape=2x2xbatch64 (native batch layout)): tenferro-eager is 10.8x faster than the slowest successful cell (jax-cpu, 0.041 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/batched_matmul_ikb_kjb_ijb (f64, threads=1, shape=4x4xbatch16 (native batch layout)): tenferro-eager is 15.7x faster than the slowest successful cell (jax-cpu, 0.042 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/batched_matmul_ikb_kjb_ijb (f64, threads=1, shape=8x8xbatch16 (native batch layout)): tenferro-eager is 12.7x faster than the slowest successful cell (jax-cpu, 0.041 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=2x2xbatch1024 (native batch layout)): jax-cpu is 16.5x faster than the slowest successful cell (tenferro-eager, 0.155 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=2x2xbatch16 (native batch layout)): jax-cpu is 20.0x faster than the slowest successful cell (tenferro-eager, 0.082 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=2x2xbatch256 (native batch layout)): jax-cpu is 18.8x faster than the slowest successful cell (tenferro-eager, 0.103 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=2x2xbatch64 (native batch layout)): jax-cpu is 18.0x faster than the slowest successful cell (tenferro-eager, 0.085 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=4x4xbatch16 (native batch layout)): jax-cpu is 15.5x faster than the slowest successful cell (tenferro-eager, 0.080 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=8x8xbatch1024 (native batch layout)): jax-cpu is 14.6x faster than the slowest successful cell (pytorch-cpu, 3.189 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=8x8xbatch16 (native batch layout)): jax-cpu is 14.5x faster than the slowest successful cell (tenferro-eager, 0.092 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=8x8xbatch256 (native batch layout)): jax-cpu is 14.4x faster than the slowest successful cell (pytorch-cpu, 0.827 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_matmul_backward (f64, threads=1, shape=8x8xbatch64 (native batch layout)): jax-cpu is 12.5x faster than the slowest successful cell (pytorch-cpu, 0.232 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_solve_backward (f64, threads=1, shape=2x2xbatch16 (native batch layout),rhs=1): jax-cpu is 15.7x faster than the slowest successful cell (tenferro-eager, 0.104 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_solve_backward (f64, threads=1, shape=2x2xbatch64 (native batch layout),rhs=1): jax-cpu is 11.4x faster than the slowest successful cell (tenferro-eager, 0.107 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_solve_backward (f64, threads=1, shape=4x4xbatch16 (native batch layout),rhs=1): jax-cpu is 17.1x faster than the slowest successful cell (tenferro-eager, 0.101 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
batched/grad_sum_batched_solve_backward (f64, threads=1, shape=8x8xbatch16 (native batch layout),rhs=1): jax-cpu is 13.0x faster than the slowest successful cell (tenferro-eager, 0.106 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/einsum_ij_jk_ik (f64, threads=1, shape=16x16): tenferro-eager is 12.9x faster than the slowest successful cell (jax-cpu, 0.041 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/einsum_ij_jk_ik (f64, threads=1, shape=2x2): tenferro-eager is 16.0x faster than the slowest successful cell (jax-cpu, 0.036 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/einsum_ij_jk_ik (f64, threads=1, shape=32x32): tenferro-eager is 12.8x faster than the slowest successful cell (jax-cpu, 0.042 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/einsum_ij_jk_ik (f64, threads=1, shape=4x4): tenferro-eager is 16.4x faster than the slowest successful cell (jax-cpu, 0.036 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/einsum_ij_jk_ik (f64, threads=1, shape=8x8): tenferro-eager is 18.1x faster than the slowest successful cell (jax-cpu, 0.041 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_lu_jvp (f64, threads=1, shape=16x16): jax-cpu is 11.6x faster than the slowest successful cell (pytorch-cpu, 0.061 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_lu_jvp (f64, threads=1, shape=2x2): jax-cpu is 13.7x faster than the slowest successful cell (pytorch-cpu, 0.050 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_lu_jvp (f64, threads=1, shape=4x4): jax-cpu is 13.5x faster than the slowest successful cell (pytorch-cpu, 0.050 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_lu_jvp (f64, threads=1, shape=8x8): jax-cpu is 11.4x faster than the slowest successful cell (pytorch-cpu, 0.057 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_lu_vjp (f64, threads=1, shape=2x2): jax-cpu is 10.8x faster than the slowest successful cell (pytorch-cpu, 0.039 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_matmul_backward (f64, threads=1, shape=16x16): jax-cpu is 18.0x faster than the slowest successful cell (tenferro-eager, 0.090 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_matmul_backward (f64, threads=1, shape=2x2): jax-cpu is 26.9x faster than the slowest successful cell (tenferro-eager, 0.101 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_matmul_backward (f64, threads=1, shape=32x32): jax-cpu is 12.3x faster than the slowest successful cell (tenferro-eager, 0.091 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_matmul_backward (f64, threads=1, shape=4x4): jax-cpu is 24.4x faster than the slowest successful cell (tenferro-eager, 0.102 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_matmul_backward (f64, threads=1, shape=8x8): jax-cpu is 16.3x faster than the slowest successful cell (tenferro-eager, 0.082 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_qr_jvp (f64, threads=1, shape=2x2): jax-cpu is 10.3x faster than the slowest successful cell (tenferro-trace, 0.039 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_qr_vjp (f64, threads=1, shape=2x2): jax-cpu is 10.9x faster than the slowest successful cell (tenferro-trace, 0.040 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_qr_vjp (f64, threads=1, shape=8x8): jax-cpu is 10.7x faster than the slowest successful cell (tenferro-trace, 0.047 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_backward (f64, threads=1, shape=16x16,rhs=1): jax-cpu is 17.8x faster than the slowest successful cell (tenferro-eager, 0.099 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_backward (f64, threads=1, shape=2x2,rhs=1): jax-cpu is 20.0x faster than the slowest successful cell (tenferro-eager, 0.096 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_backward (f64, threads=1, shape=32x32,rhs=1): jax-cpu is 14.3x faster than the slowest successful cell (tenferro-eager, 0.113 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_backward (f64, threads=1, shape=4x4,rhs=1): jax-cpu is 20.7x faster than the slowest successful cell (tenferro-eager, 0.097 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_backward (f64, threads=1, shape=8x8,rhs=1): jax-cpu is 19.9x faster than the slowest successful cell (tenferro-eager, 0.097 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_jvp (f64, threads=1, shape=16x16,rhs=1): jax-cpu is 17.0x faster than the slowest successful cell (pytorch-cpu, 0.082 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_jvp (f64, threads=1, shape=2x2,rhs=1): jax-cpu is 22.4x faster than the slowest successful cell (pytorch-cpu, 0.081 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_jvp (f64, threads=1, shape=32x32,rhs=1): jax-cpu is 12.7x faster than the slowest successful cell (pytorch-cpu, 0.090 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_jvp (f64, threads=1, shape=4x4,rhs=1): jax-cpu is 22.1x faster than the slowest successful cell (pytorch-cpu, 0.081 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_solve_jvp (f64, threads=1, shape=8x8,rhs=1): jax-cpu is 21.3x faster than the slowest successful cell (pytorch-cpu, 0.081 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_svd_s_backward (f64, threads=1, shape=2x2): jax-cpu is 16.6x faster than the slowest successful cell (tenferro-eager, 0.082 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.
small/grad_sum_svd_s_backward (f64, threads=1, shape=4x4): jax-cpu is 14.5x faster than the slowest successful cell (tenferro-eager, 0.085 ms). Audit fixture semantics, synchronization, labels, and operation-specific bandwidth/FLOP bounds.

Physical Bound Audit

This independent check derives minimum logical bytes from dtype and shape for materializing public API operations, plus conservative operation FLOPs for reductions, Frobenius norms, matrix products, triangular solves, and recognized dense linalg families. Decorated shapes (->, rhs=, and +) are parsed explicitly. Operations whose touched region cannot be inferred conservatively from the reported shape are left unmodeled. It assumes at most 100 GB/s and 50 GFLOP/s per requested CPU thread, capped at 1000 GB/s and 3200 GFLOP/s. A cell more than 10x faster than the resulting lower bound is flagged. These deliberately generous ceilings are a sanity screen, not a performance gate.

No cell exceeded the conservative physical bound by more than 10x.

Threads: 4

Source report: data/results/mac-cpu/cpu/einsum/20260923_204226/cpu_ops_report.md.

Suite: cpu/cpu_ops
Target profile: mac-cpu
Timestamp: 20260923_204226

Latest run: ./scripts/run_all.sh 4.

This file is generated from one CPU ops run under data/results/mac-cpu/cpu/einsum/20260923_204226.

tenferro-rs commit: 8a1839febeb3c868a502e26397ca10761bbc568d