• Global
  • Configurations
• Summary
• Functions
• Loops

Help is available by moving the cursor above any symbol or by checking MAQAO website.

Global Metrics

Metric		r0	r1	r2
Total Time (s)		3.00	3.40	3.48
Max (Thread Active Time) (s)		1.13	1.49	1.69
Average Active Time (s)		0.43	0.27	0.27
Activity Ratio (%)		53.0	33.2	33.4
Average number of active threads		9.094	5.020	5.007
Affinity Stability (%)		62.2	39.3	39.0
Time in analyzed loops (%)		53.0	85.7	85.5
Time in analyzed innermost loops (%)		51.5	80.9	78.2
Time in user code (%)		53.6	86.5	86.2
Compilation Options Score (%)		99.2	75.0	100
Array Access Efficiency (%)		72.7	73.1	74.1
Potential Speedups
Perfect Flow Complexity		1.00	1.01	1.00
Perfect OpenMP/MPI/Pthread/TBB		1.02	1.02	1.01
Perfect OpenMP/MPI/Pthread/TBB + Perfect Load Distribution		4.81	6.30	7.00
No Scalar Integer	Potential Speedup	1.01	1.01	1.01
	Nb Loops to get 80%	1	2	5
FP Vectorised	Potential Speedup	1.09	1.15	1.14
	Nb Loops to get 80%	1	1	1
Fully Vectorised	Potential Speedup	1.65	2.87	2.91
	Nb Loops to get 80%	1	2	3
Only FP Arithmetic	Potential Speedup	1.70	2.89	3.10
	Nb Loops to get 80%	1	2	3

Cumulated Speedup If No Scalar Integer

Cumulated Speedup If FP Vectorized

Cumulated Speedup If Fully Vectorized

Cumulated Speedup If Only FP Arithmetic

Loop Based Profiles

Innermost / Single Loops

Inbetween Loops

Outermost Loops

Cumulated Coverage With All Loops

Innermost Loop Based Profiles

Coverage

Count

Application Categorization

Time

Coverage

Compilation Options

r0r1r2

Source Object	Issue
▼libllama.so–
▼llama-vocab.cpp–
○	-mcpu=native is missing.
▼llama-arch.cpp–
○	-mcpu=native is missing.
▼hashtable_policy.h–
○	-mcpu=native is missing.
▼unicode.cpp–
○	-mcpu=native is missing.
▼hashtable.h–
○	-mcpu=native is missing.
▼libggml-cpu.so–
▼binary-ops.cpp–
○
▼traits.cpp–
○
▼kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c–
○
▼kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p4x8_16x4_neon_i8mm.c–
○
▼kleidiai.cpp–
○
▼vec.cpp–
○
▼kai_lhs_quant_pack_qsi8d32p4x8sb_f32_neon.c–
○
▼ops.cpp–
○
▼ggml-cpu.c–
○
▼quants.c–
○
▼[vdso]–
▼–
○	-g is missing for some functions (possibly ones added by the compiler), it is needed to have more accurate reports. Other recommended flags are: -O2/-O3, -march=(target)
○	-O2, -O3 or -Ofast is missing.
○	-mcpu=native is missing.
▼libggml-base.so–
▼ggml-quants.c–
○	-mcpu=native is missing.
▼ggml.c–
○	-mcpu=native is missing.
▼stl_construct.h–
○	-mcpu=native is missing.

Source Object	Issue
▼libllama.so–
▼llama-vocab.cpp–
○	-funroll-loops is missing.
▼basic_string.tcc–
○	-funroll-loops is missing.
▼llama-impl.cpp–
○	-funroll-loops is missing.
▼unicode.cpp–
○	-funroll-loops is missing.
▼hashtable.h–
○	-funroll-loops is missing.
▼libggml-cpu.so–
▼vec.cpp–
○	-funroll-loops is missing.
▼kernels.cpp–
○	-funroll-loops is missing.
▼kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c–
○	-funroll-loops is missing.
▼invoke.h–
○	-funroll-loops is missing.
▼variant–
○	-funroll-loops is missing.
▼ops.cpp–
○	-funroll-loops is missing.
▼kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p4x8_16x4_neon_i8mm.c–
○	-funroll-loops is missing.
▼kai_lhs_quant_pack_qsi8d32p4x8sb_f32_neon.c–
○	-funroll-loops is missing.
▼binary-ops.cpp–
○	-funroll-loops is missing.
▼ggml-cpu.c–
○	-funroll-loops is missing.
▼quants.c–
○	-funroll-loops is missing.
▼libggml-base.so–
▼ggml.c–
○	-funroll-loops is missing.
▼basic_string.tcc–
○	-funroll-loops is missing.
▼stl_vector.h–
○	-funroll-loops is missing.

Source Object	Issue
▼libllama.so–
▼llama-vocab.cpp–
○
▼llama-impl.cpp–
○
▼hashtable.h–
○
▼hashtable_policy.h–
○
▼libggml-cpu.so–
▼binary-ops.cpp–
○
▼traits.cpp–
○
▼kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c–
○
▼kai_matmul_clamp_f32_qsi8d32p4x8_qsi4c32p4x8_16x4_neon_i8mm.c–
○
▼kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p4x8_1x4x32_neon_dotprod.c–
○
▼kleidiai.cpp–
○
▼vec.cpp–
○
▼kai_lhs_quant_pack_qsi8d32p4x8sb_f32_neon.c–
○
▼ops.cpp–
○
▼ggml-cpu.c–
○
▼quants.c–
○
▼libggml-base.so–
▼gguf.cpp–
○
▼basic_string.tcc–
○
▼ggml.c–
○

Path Count Profiles

Coverage

Count

Low Iteration Count Profiles

Coverage

Count

Average Number of Active Threads

Run 1 - orig_default

Run 2 - gcc_default

Run 3 - gcc_4