Tensor timing report card

[mpharrigan]

Not all bloq examples support tensor simulation. This report card automatically determines which bloq examples should be tensor simulable.

• State vector simulation uses $2^n$ numbers to simulate a quantum state. The tensor protocol uses quimb to try to find more efficient contraction orderings. Quimb reports the contraction width, which is the minimum size of any intermediate tensor encountered in the contraciton. The simulation uses $2^w$ numbers, where $w$ is the width. We consider a width under 25 qubits as simulable.
• Qualtran requires "flattening" out the bloq to turn it into an efficient tensor network. This may take too much time itself for large algorithms with many levels of abstraction. If the process of turning a bloq into a quimb tensor network and finding a contraction ordering takes longer than 8 seconds, we don't consider the bloq simulable.
• The flattened structure needs to have explicit tensors. For bloqs with symbolic parameters, we either can't decompose & flatten them, or the tensors would be symbolic, which we don't support.

[mpharrigan]

Number of bloq examples considered

print(len(df))

304

Number of bloq examples successfully flattened

print(len(df[df['flat_dur'] > 0]))

263

Number of bloq examples with tensors

print(len(df[df['width'] > 0]))

157

Bloqs that are tensor simulable

print(len(df[df['width'] <= 25]))
df[df['width'] <= 25]

120

	name	cls	flat_dur	err	tn_dur	width	width_dur
1	add_k_small	AddK	0.026146	NaN	0.003073	8.0	0.003242
4	add_oop_small	OutOfPlaceAdder	0.027389	NaN	0.003113	21.0	0.003324
8	add_small	Add	0.023498	NaN	0.002645	16.0	0.002721
10	bitwise_not	BitwiseNot	0.001586	NaN	0.000363	8.0	0.000347
12	xor	Xor	0.006454	NaN	0.000576	16.0	0.000651
...	...	...	...	...	...	...	...
289	sparse_state_prep_via_rotations	SparseStatePreparationViaRotations	1.489863	NaN	0.044312	12.0	0.039386
294	state_prep_alias	StatePreparationAliasSampling	0.178420	NaN	0.010980	22.0	0.012333
296	state_prep_via_rotation	StatePreparationViaRotations	0.311990	NaN	0.014321	12.0	0.009816
297	approx_cswap_small	CSwapApprox	0.037106	NaN	0.003697	18.0	0.003618
300	swz_small	SwapWithZero	0.019082	NaN	0.002103	14.0	0.001731

120 rows × 7 columns

Bloqs whose tensor network is too big

df[df['width'] > 25].sort_values(by='width')

	name	cls	flat_dur	err	tn_dur	width	width_dur
20	leq	LessThanEqual	0.173596	NaN	0.014591	2.600000e+01	0.017194
153	cplx_givens	ComplexGivensRotationByPhaseGradient	0.135327	NaN	0.010783	2.800000e+01	0.012214
51	comparator	Comparator	0.454121	NaN	0.023943	2.900000e+01	0.025806
299	multiplexed_cswap	MultiplexedCSwap	0.116693	NaN	0.008720	3.000000e+01	0.008778
258	prepare_identity	PrepareIdentity	0.001847	NaN	0.000339	3.000000e+01	0.000447
234	kaiser_window_state_small	KaiserWindowState	0.000314	NaN	0.180226	3.200000e+01	0.000075
291	sparse_state_prep_alias_from_list	SparseStatePreparationAliasSampling	0.309500	NaN	0.057082	3.400000e+01	0.024256
290	sparse_state_prep_alias	SparseStatePreparationAliasSampling	0.244710	NaN	0.013807	3.400000e+01	0.014747
25	ctrl_add_or_sub_signed	ControlledAddOrSubtract	0.076607	NaN	0.009669	3.400000e+01	0.009102
27	ctrl_add_or_sub_unsigned	ControlledAddOrSubtract	0.080349	NaN	0.007428	3.400000e+01	0.008773
105	scaled_chebyshev_poly_odd	ScaledChebyshevPolynomial	2.744674	NaN	0.121746	3.600000e+01	0.059081
233	black_box_select	BlackBoxSelect	0.267093	NaN	0.019364	3.700000e+01	0.020489
56	sub_diff_size_regs	Subtract	0.265051	NaN	0.033848	4.000000e+01	0.047651
184	qrom_multi_dim	QROM	0.127911	NaN	0.010900	4.000000e+01	0.012139
2	add_diff_size_regs	Add	0.125302	NaN	0.010692	4.000000e+01	0.016339
55	parallel_compare	ParallelComparators	0.631275	NaN	0.025924	4.500000e+01	0.028546
212	ctrl_on_multiple_values	CtrlSpecAnd	0.077075	NaN	0.008729	4.700000e+01	0.009076
302	swz_multi_dimensional	SwapWithZero	0.440719	NaN	0.021182	5.600000e+01	0.020162
108	select_block	SelectBlockEncoding	1.075520	NaN	0.107634	6.200000e+01	0.048434
104	black_box_select_block	SelectBlockEncoding	0.860905	NaN	0.110791	6.200000e+01	0.049564
182	qroam_clean_multi_dim	QROAMClean	2.641734	NaN	0.093979	6.300000e+01	0.079423
224	moddbl_large	ModDbl	2.568159	NaN	0.070820	6.400000e+01	0.088628
127	cast	Cast	0.000413	NaN	0.000515	6.400000e+01	0.000750
208	controlled_via_and_ints	ControlledViaAnd	0.284635	NaN	0.022803	6.600000e+01	0.024899
188	qroam_multi_dim	SelectSwapQROM	1.952651	NaN	0.061592	6.800000e+01	0.086867
140	sel_hubb	SelectHubbard	2.580288	NaN	0.160665	8.800000e+01	0.184910
130	partition	Partition	0.000736	NaN	0.000838	9.800000e+01	0.001213
3	add_k_large	AddK	2.457865	NaN	0.059918	1.280000e+02	0.074924
61	sub_large	Subtract	2.606371	NaN	0.070517	2.560000e+02	0.090522
59	sub_from_large	SubtractFrom	2.378148	NaN	0.061737	2.560000e+02	0.072526
6	add_large	Add	1.964227	NaN	0.051067	2.560000e+02	0.063128
83	cswap_large	CSwap	1.041534	NaN	0.041354	2.580000e+02	0.122483
301	approx_cswap_large	CSwapApprox	3.643124	NaN	0.056535	2.580000e+02	0.065362
5	add_oop_large	OutOfPlaceAdder	1.619623	NaN	0.046654	3.210000e+02	0.057121
36	plus_equal_product	PlusEqualProduct	0.000766	NaN	0.002239	6.553600e+04	0.000107
265	add_into_phase_grad	AddIntoPhaseGrad	0.000497	NaN	0.004727	6.553600e+04	0.000104
39	square	Square	0.000426	NaN	0.001193	4.294967e+09	0.000140

Bloqs without tensors

Due to errors encountered in flattening or if the bloq's callees don't support tensor simulation.

df[df['width'].isna()]

	name	cls	flat_dur	err	tn_dur	width	width_dur
0	add_k	AddK	0.016689	AddK does not support tensor simulation.	NaN	NaN	NaN
7	add_oop_symb	OutOfPlaceAdder	0.017955	OutOfPlaceAdder does not support tensor simula...	NaN	NaN	NaN
9	add_symb	Add	0.018177	Add does not support tensor simulation.	NaN	NaN	NaN
11	bitwise_not_symb	BitwiseNot	0.016773	XGate⨂n does not support tensor simulation.	NaN	NaN	NaN
13	xor_symb	Xor	0.021624	Xor does not support tensor simulation.	NaN	NaN	NaN
...	...	...	...	...	...	...	...
292	sparse_state_prep_alias_symb	SparseStatePreparationAliasSampling	0.005134	PrepareUniformSuperposition does not support t...	NaN	NaN	NaN
293	state_prep_alias_symb	StatePreparationAliasSampling	0.018301	PrepareUniformSuperposition does not support t...	NaN	NaN	NaN
295	state_prep_via_rotation_symb	StatePreparationViaRotations	NaN	Split cannot have a symbolic data type.	NaN	NaN	NaN
298	approx_cswap_symb	CSwapApprox	0.016267	CSwapApprox does not support tensor simulation.	NaN	NaN	NaN
303	swz	SwapWithZero	NaN	Timeout	NaN	NaN	NaN

147 rows × 7 columns

Slowest to flatten

Within the overall timeout

df.sort_values(by='flat_dur', ascending=False).head()

	name	cls	flat_dur	err	tn_dur	width	width_dur
107	black_box_lcu_block	LCUBlockEncoding	4.766397	ModAddK does not support tensor simulation.	NaN	NaN	NaN
301	approx_cswap_large	CSwapApprox	3.643124	NaN	0.056535	258.0	0.065362
105	scaled_chebyshev_poly_odd	ScaledChebyshevPolynomial	2.744674	NaN	0.121746	36.0	0.059081
182	qroam_clean_multi_dim	QROAMClean	2.641734	NaN	0.093979	63.0	0.079423
61	sub_large	Subtract	2.606371	NaN	0.070517	256.0	0.090522

Flattening is the rate-limiting step.

For bloqs that have been successfully flattened, the maximum tensor-network-construction and tensor-contraction-ordering durations are less than 0.5s. Note: the contraction finding code uses the fast, naive approach. One can choose more expensive approaches where the contraciton-ordering-finding is more expensive.

# Slowest tn_dur
df.sort_values(by='tn_dur', ascending=False).head()

	name	cls	flat_dur	err	tn_dur	width	width_dur
234	kaiser_window_state_small	KaiserWindowState	0.000314	NaN	0.180226	32.0	0.000075
140	sel_hubb	SelectHubbard	2.580288	NaN	0.160665	88.0	0.184910
105	scaled_chebyshev_poly_odd	ScaledChebyshevPolynomial	2.744674	NaN	0.121746	36.0	0.059081
104	black_box_select_block	SelectBlockEncoding	0.860905	NaN	0.110791	62.0	0.049564
108	select_block	SelectBlockEncoding	1.075520	NaN	0.107634	62.0	0.048434

# Slowest width_dur
df.sort_values(by='width_dur', ascending=False).head()

	name	cls	flat_dur	err	tn_dur	width	width_dur
140	sel_hubb	SelectHubbard	2.580288	NaN	0.160665	88.0	0.184910
83	cswap_large	CSwap	1.041534	NaN	0.041354	258.0	0.122483
49	sparse_permutation	Permutation	0.881599	NaN	0.031373	12.0	0.105166
16	clineardepthgreaterthan_example	CLinearDepthGreaterThan	0.224516	NaN	0.011326	24.0	0.098919
103	chebyshev_poly_even	ChebyshevPolynomial	0.978074	NaN	0.027787	15.0	0.092611