options

Loops Index

40 loops have been discarded from the report because their ratio ((Max Inclusive Time Over Threads * 100) / Max Thread Active Time) is lower than the threshold set by object_coverage_threshold (0.1%). It represents about 0.05% of the application. To include them, change the value of object_coverage_threshold in the experiment directory configuration file, then rerun the command with the additionnal parameter --force-static-analysis

Columns Filter

Level Max Thread Time / Walltime aocc_5 (%) Exclusive Coverage aocc_5 (%) Inclusive Coverage aocc_5 (%) Max Exclusive Time Over Threads aocc_5 (s) Max Inclusive Time Over Threads aocc_5 (s) Exclusive Time w.r.t. Wall Time aocc_5 (s) Inclusive Time w.r.t. Wall Time aocc_5 (s) Nb Threads aocc_5 GFLOPS aocc_5 Vectorization Ratio (%) Vector Length Use (%) Speedup If No Scalar Integer Speedup If FP Vectorized Speedup If Fully Vectorized Speedup If Perfect Load Balancing aocc_5 Stride 0 Stride 1 Stride n Stride Unknown Stride Indirect Array Access Efficiency Level Max Thread Time / Walltime Exclusive Coverage Inclusive Coverage Max Exclusive Time Over Threads Max Inclusive Time Over Threads Exclusive Time w.r.t. Wall Time Inclusive Time w.r.t. Wall Time Nb Threads GFLOPS Vectorization Ratio Vector Length Use Speedup If No Scalar Integer Speedup If FP Vectorized Speedup If Fully Vectorized Speedup If Perfect Load Balancing Stride 0 Stride 1 Stride n Stride Unknown Stride Indirect Array Access Efficiency
Loop idSource LocationSource FunctionLevelMax Thread Time / Walltime aocc_5 (%)Exclusive Coverage aocc_5 (%)Inclusive Coverage aocc_5 (%)Max Exclusive Time Over Threads aocc_5 (s)Max Inclusive Time Over Threads aocc_5 (s)Exclusive Time w.r.t. Wall Time aocc_5 (s)Inclusive Time w.r.t. Wall Time aocc_5 (s)Nb Threads aocc_5GFLOPS aocc_5Vectorization Ratio (%)Vector Length Use (%)Speedup If No Scalar IntegerSpeedup If FP VectorizedSpeedup If Fully VectorizedSpeedup If Perfect Load Balancing aocc_5Stride 0Stride 1Stride nStride UnknownStride IndirectArray Access Efficiency
97libggml-cpu.so - ggml-cpu.c:1291-1297ggml_compute_forward_mul_matInnermost1.791.651.650.540.540.240.241891.80011.38112.462.23NANANANANA0.00
3065libggml-cpu.so - sgemm.cpp:144-464 [...]void (anonymous namespace)::tinyBLAS<16, float __vector(16), float __vector(16), unsigned short, unsigned short, float>::gemm<4, 6, 2>(long, long, long)Innermost0.400.500.500.120.120.070.071923018.37NANANANANA1.66NANANANANA0.00
4libggml-cpu.so - ggml-impl.h:346-404 [...]ggml_cpu_fp32_to_fp16Single0.300.180.180.090.090.030.031602.219.098.521.463.7513.882.9502000100.00
2078libggml-cpu.so - vec.h:89-89ggml_compute_forward_soft_maxInnermost0.230.170.170.070.070.020.0219215.13100501122.8502000100.00
386libggml-cpu.so - mmq.cpp:303-1392 [...]void parallel_for<(anonymous namespace)::convert_B_packed_format<block_q8_0, 32>(void*, block_q8_0 const*, int, int)::{lambda(int, int)#1}>(int, (anonymous namespace)::convert_B_packed_format<block_q8_0, 32>(void*, block_q8_0 const*, int,...Innermost0.180.130.130.050.050.020.021690.0090.9138.761.4711.412.6423009085.94
1299libggml-cpu.so - vec.h:1084-1115 [...]ggml_vec_swiglu_f32Single0.170.100.100.050.050.010.011711336.359898.131113.040.5003056.25
×