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	0.99 🟡	1.02 ✅	1.02 ✅	0.55 🟡	0.70 🟡	0.70 🟡	0.72 🟡	0.74 🟡
1M	0.95 🟡	0.94 🟡	0.95 🟡	0.65 🟡	0.87 🟡	0.73 🟡	0.75 🟡	0.68 🟡

Geomean by dt

size	f64	f32	c128	dec	f16	i32	i64
100K	0.87 🟡	1.00 🟡	0.98 🟡	0.08 🔴	1.03 ✅	1.33 ✅	1.14 ✅
1M	1.12 ✅	1.15 ✅	1.04 ✅	0.07 🔴	1.02 ✅	1.19 ✅	1.02 ✅

Geomean by op

size	sum	min	max	prod
100K	0.80 🟡	0.66 🟡	0.66 🟡	1.19 ✅
1M	0.84 🟡	0.65 🟡	0.64 🟡	1.28 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
1M|dec|bcast|sum|ax0	5.5269	0.0655	0.01 🔴
100K|dec|bcast|sum|ax0	0.5555	0.0087	0.02 🔴
100K|dec|F|sum|ax1	0.5585	0.0094	0.02 🔴
100K|dec|C|sum|ax0	0.5615	0.0111	0.02 🔴
100K|dec|T|sum|ax1	0.5582	0.0111	0.02 🔴
100K|dec|negrow|sum|ax0	0.5592	0.0114	0.02 🔴
100K|dec|sliced|sum|ax0	0.5517	0.0115	0.02 🔴
1M|dec|sliced|sum|ax0	5.5802	0.1240	0.02 🔴
1M|dec|negrow|sum|ax0	5.5707	0.1242	0.02 🔴
1M|dec|F|sum|ax1	5.6157	0.1254	0.02 🔴
1M|dec|T|sum|ax1	5.5849	0.1278	0.02 🔴
1M|dec|C|sum|ax0	5.5736	0.1524	0.03 🔴
100K|dec|C|min|ax0	0.3021	0.0141	0.05 🔴
1M|dec|negcol|sum|ax0	5.6678	0.2866	0.05 🔴
100K|dec|negcol|sum|ax0	0.5594	0.0302	0.05 🔴

Copy / identity-ufunc (np.positive)

Geomean by lay

size	C	F	T	strided	sliced	negrow	negcol	bcast
100K	1.22 ✅	1.51 ✅	1.28 ✅	0.83 🟡	1.54 ✅	1.48 ✅	2.20 ✅	1.62 ✅
1M	2.82 ✅	2.89 ✅	2.88 ✅	2.08 ✅	2.58 ✅	2.64 ✅	3.16 ✅	2.66 ✅

Geomean by dt

size	u8	i8	i16	u16	i32	u32	i64	u64	char	f16	f32	f64	c128
100K	0.92 🟡	1.03 ✅	1.51 ✅	1.99 ✅	1.02 ✅	1.09 ✅	1.65 ✅	1.08 ✅	2.56 ✅	2.47 ✅	1.15 ✅	0.85 🟡	2.56 ✅
1M	4.60 ✅	4.87 ✅	2.13 ✅	2.14 ✅	2.11 ✅	2.12 ✅	2.56 ✅	2.46 ✅	2.20 ✅	2.10 ✅	2.14 ✅	2.59 ✅	5.47 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
100K|f64|strided|pos	0.0243	0.0100	0.41 🟠
100K|u64|strided|pos	0.0241	0.0110	0.46 🟠
100K|i64|strided|pos	0.0237	0.0109	0.46 🟠
100K|f32|strided|pos	0.0216	0.0100	0.46 🟠
100K|i32|strided|pos	0.0213	0.0110	0.52 🟡
100K|i64|C|pos	0.0400	0.0210	0.53 🟡
100K|i64|T|pos	0.0387	0.0208	0.54 🟡
100K|i8|sliced|pos	0.0423	0.0228	0.54 🟡
100K|u8|bcast|pos	0.0416	0.0229	0.55 🟡
100K|u8|sliced|pos	0.0410	0.0227	0.55 🟡
100K|u8|negrow|pos	0.0410	0.0231	0.56 🟡
100K|u32|strided|pos	0.0191	0.0109	0.57 🟡
100K|f64|T|pos	0.0389	0.0227	0.58 🟡
100K|f64|F|pos	0.0362	0.0227	0.63 🟡
100K|f64|C|pos	0.0361	0.0228	0.63 🟡

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

Geomean by lay

size	C	F	T	strided	sliced	negrow	negcol	bcast
100K	0.73 🟡	0.70 🟡	0.68 🟡	0.53 🟡	0.84 🟡	0.85 🟡	1.18 ✅	0.85 🟡
1M	1.54 ✅	1.48 ✅	1.48 ✅	1.12 ✅	1.64 ✅	1.65 ✅	1.80 ✅	1.63 ✅

Geomean by dt

size	f64	f32	c128	f16	i32	i64
100K	0.67 🟡	0.57 🟡	1.15 ✅	0.75 🟡	0.82 🟡	0.81 🟡
1M	1.86 ✅	1.53 ✅	1.73 ✅	0.94 🟡	1.54 ✅	1.79 ✅

Geomean by op

size	add	mul	neg	abs	sqrt	less	copy
100K	1.04 ✅	0.99 🟡	0.76 🟡	0.74 🟡	0.94 🟡	0.61 🟡	0.51 🟡
1M	1.81 ✅	1.78 ✅	2.11 ✅	1.62 ✅	1.50 ✅	0.68 🟡	1.75 ✅

Worst 15 cells (NumSharp slowest vs NumPy)

key	NumSharp ms	NumPy ms	ratio
100K|f64|strided|abs	0.0466	0.0077	0.17 🔴
100K|f64|strided|neg	0.0473	0.0088	0.19 🔴
100K|i32|bcast|copy	0.0233	0.0050	0.21 🟠
100K|f16|bcast|copy	0.0134	0.0030	0.22 🟠
100K|f16|negrow|copy	0.0168	0.0038	0.23 🟠
100K|f32|C|copy	0.0244	0.0059	0.24 🟠
100K|i64|strided|neg	0.0296	0.0076	0.26 🟠
100K|f64|strided|copy	0.0413	0.0108	0.26 🟠
100K|f64|negrow|copy	0.0595	0.0160	0.27 🟠
100K|f16|sliced|copy	0.0134	0.0037	0.27 🟠
100K|f64|sliced|copy	0.0592	0.0162	0.27 🟠
100K|i32|negrow|copy	0.0234	0.0064	0.28 🟠
100K|f32|C|add	0.0240	0.0066	0.28 🟠
100K|f64|strided|mul	0.0509	0.0142	0.28 🟠
100K|f64|strided|add	0.0516	0.0147	0.28 🟠