Layout suite — reduction / copy / elementwise × memory layout × dtype

ratio = NumPy_ms / NumSharp_ms — >1.0 = NumSharp faster. ✅≥1.0 🟡≥0.5 🟠≥0.2 🔴<0.2. Layouts (8, harmonized with the cast subsystem): C, F (Fortran), T (transpose), strided [:, ::2], sliced (offset), negrow [::-1,:], negcol [:,::-1], bcast (stride-0). Fills the op-matrix's blind spot (it measures C-contiguous only). 100K + 1M elements, best-of-rounds.

Reduction (sum/min/max/prod, both axes)

Geomean by lay

size	C	F	T	strided	negrow	negcol	sliced	bcast
100K	1.10 ✅	1.11 ✅	1.10 ✅	0.60 🟡	1.03 ✅	0.73 🟡	0.91 🟡	0.83 🟡
1M	1.05 ✅	1.01 ✅	0.98 🟡	0.63 🟡	1.02 ✅	0.72 🟡	0.81 🟡	0.75 🟡

Geomean by dt

size	f64	f32	c128	dec	f16	i32	i64
100K	1.11 ✅	1.30 ✅	0.97 🟡	0.12 🔴	1.05 ✅	1.50 ✅	1.14 ✅
1M	1.02 ✅	1.28 ✅	0.96 🟡	0.11 🔴	1.07 ✅	1.31 ✅	1.07 ✅

Geomean by op

size	sum	min	max	prod
100K	0.95 🟡	0.74 🟡	0.76 🟡	1.34 ✅
1M	0.95 🟡	0.67 🟡	0.68 🟡	1.32 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
1M|dec|bcast|sum|ax0	7.3695	0.1408	0.02 🔴
1M|dec|bcast|sum|ax1	4.1086	0.1125	0.03 🔴
100K|dec|bcast|sum|ax0	0.7283	0.0204	0.03 🔴
1M|dec|T|sum|ax1	7.5294	0.2240	0.03 🔴
1M|dec|F|sum|ax1	7.5579	0.2255	0.03 🔴
100K|dec|C|sum|ax0	0.7326	0.0227	0.03 🔴
100K|dec|F|sum|ax1	0.7286	0.0227	0.03 🔴
100K|dec|T|sum|ax1	0.7284	0.0227	0.03 🔴
100K|dec|negrow|sum|ax0	0.7282	0.0230	0.03 🔴
1M|dec|negrow|sum|ax0	7.4363	0.2417	0.03 🔴
1M|dec|sliced|sum|ax0	7.2451	0.2436	0.03 🔴
100K|dec|sliced|sum|ax0	0.7190	0.0248	0.03 🔴
1M|dec|C|sum|ax0	5.5333	0.2184	0.04 🔴
100K|dec|negrow|sum|ax1	0.4159	0.0184	0.04 🔴
1M|i32|bcast|max|ax1	1.1684	0.0603	0.05 🔴

Copy / identity-ufunc (np.positive)

Geomean by lay

size	C	F	T	strided	sliced	negrow	negcol	bcast
100K	3.78 ✅	2.96 ✅	3.47 ✅	1.55 ✅	3.76 ✅	3.95 ✅	3.58 ✅	4.26 ✅
1M	6.83 ✅	6.29 ✅	6.51 ✅	3.32 ✅	5.41 ✅	6.04 ✅	6.47 ✅	9.24 ✅

Geomean by dt

size	u8	i8	i16	u16	i32	u32	i64	u64	char	f16	f32	f64	c128
100K	5.43 ✅	4.13 ✅	4.31 ✅	4.58 ✅	2.80 ✅	2.76 ✅	1.87 ✅	1.51 ✅	4.72 ✅	4.89 ✅	1.73 ✅	1.28 ✅	10.68 ✅
1M	10.08 ✅	10.08 ✅	8.28 ✅	8.35 ✅	4.46 ✅	4.98 ✅	3.65 ✅	4.39 ✅	7.96 ✅	7.65 ✅	4.17 ✅	3.93 ✅	5.93 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
100K|f32|strided|pos	0.0285	0.0159	0.56 🟡
100K|f64|strided|pos	0.0312	0.0180	0.58 🟡
100K|i32|strided|pos	0.0286	0.0197	0.69 🟡
100K|u32|strided|pos	0.0285	0.0197	0.69 🟡
100K|f64|T|pos	0.0310	0.0277	0.89 🟡
100K|i64|strided|pos	0.0317	0.0299	0.94 🟡
100K|u64|strided|pos	0.0312	0.0298	0.96 🟡
100K|f64|C|pos	0.0281	0.0277	0.98 🟡
100K|i64|F|pos	0.0370	0.0364	0.98 🟡
100K|f64|F|pos	0.0305	0.0329	1.08 ✅
100K|u64|F|pos	0.0317	0.0364	1.15 ✅
100K|u64|T|pos	0.0310	0.0365	1.18 ✅
100K|u64|C|pos	0.0284	0.0364	1.28 ✅
1M|i32|strided|pos	0.2711	0.3486	1.29 ✅
100K|i8|strided|pos	0.0089	0.0117	1.31 ✅

Elementwise (add/mul/neg/abs/sqrt/less/copy)

Geomean by lay

size	C	F	T	strided	sliced	negrow	negcol	bcast
100K	1.26 ✅	1.19 ✅	1.15 ✅	0.54 🟡	1.59 ✅	1.55 ✅	1.13 ✅	1.46 ✅
1M	2.15 ✅	1.96 ✅	1.97 ✅	0.99 🟡	2.38 ✅	2.40 ✅	1.49 ✅	2.89 ✅

Geomean by dt

size	f64	f32	c128	f16	i32	i64
100K	1.07 ✅	0.86 🟡	2.61 ✅	0.91 🟡	1.27 ✅	0.98 🟡
1M	2.14 ✅	2.34 ✅	1.93 ✅	1.12 ✅	2.40 ✅	2.05 ✅

Geomean by op

size	add	mul	neg	abs	sqrt	less	copy
100K	1.44 ✅	1.39 ✅	1.30 ✅	1.20 ✅	0.93 🟡	0.87 🟡	1.26 ✅
1M	2.18 ✅	2.17 ✅	2.80 ✅	2.44 ✅	1.27 ✅	0.88 🟡	2.88 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
100K|i64|strided|add	0.1828	0.0256	0.14 🔴
100K|f64|strided|add	0.1900	0.0287	0.15 🔴
100K|f64|strided|mul	0.1833	0.0301	0.16 🔴
100K|i64|strided|mul	0.1831	0.0301	0.16 🔴
100K|f64|strided|neg	0.0943	0.0163	0.17 🔴
100K|f64|strided|abs	0.0931	0.0165	0.18 🔴
100K|f64|T|sqrt	0.2797	0.0552	0.20 🔴
100K|f32|strided|abs	0.0492	0.0098	0.20 🟠
100K|i64|strided|sqrt	0.1752	0.0408	0.23 🟠
100K|i32|bcast|sqrt	0.3200	0.0804	0.25 🟠
100K|f32|negcol|add	0.1845	0.0494	0.27 🟠
100K|i32|F|sqrt	0.2797	0.0778	0.28 🟠
100K|i32|strided|add	0.0932	0.0279	0.30 🟠
100K|i32|strided|mul	0.0930	0.0279	0.30 🟠
100K|f32|strided|add	0.0933	0.0286	0.31 🟠