QSimSimulator ignores initial_state when not using AVX2
Closed this issue · 4 comments
maffoo commented
The following does not work on a machine without avx2 support:
In [1]: import cirq; import qsimcirq
In [2]: q = cirq.LineQubit(0)
In [3]: c = cirq.Circuit(cirq.X(q)**0.25)
In [4]: s = qsimcirq.QSimSimulator()
In [5]: r = s.simulate(c, initial_state=0); r
Out[5]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩
In [6]: r = s.simulate(c, initial_state=r.state_vector()); r
Out[6]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩
In [7]: r = s.simulate(c, initial_state=r.state_vector()); r
Out[7]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩
In [8]: r = s.simulate(c, initial_state=r.state_vector()); r
Out[8]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩Here's the output of lscpu:
$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 46 bits physical, 48 bits virtual
CPU(s): 12
On-line CPU(s) list: 0-11
Thread(s) per core: 2
Core(s) per socket: 6
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 62
Model name: Intel(R) Xeon(R) CPU E5-1650 v2 @ 3.50GHz
Stepping: 4
CPU MHz: 3491.789
CPU max MHz: 3900.0000
CPU min MHz: 1200.0000
BogoMIPS: 6983.57
Virtualization: VT-x
L1d cache: 192 KiB
L1i cache: 192 KiB
L2 cache: 1.5 MiB
L3 cache: 12 MiB
NUMA node0 CPU(s): 0-11
Vulnerability Itlb multihit: KVM: Mitigation: VMX disabled
Vulnerability L1tf: Mitigation; PTE Inversion; VMX conditional cache flushes, SMT vulnerable
Vulnerability Mds: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Meltdown: Mitigation; PTI
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Full generic retpoline, IBPB conditional, IBRS_FW, STIBP con
ditional, RSB filling
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36
clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtsc
p lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_
tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2
ssse3 cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic popcnt tsc_deadline_
timer aes xsave avx f16c rdrand lahf_lm cpuid_fault epb pti ssbd ibrs ib
pb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveo
pt dtherm ida arat pln pts md_clear flush_l1d
This works as expected on an avx2-capable machine:
In [1]: import cirq; import qsimcirq
In [2]: q = cirq.LineQubit(0)
In [3]: c = cirq.Circuit(cirq.X(q)**0.25)
In [4]: s = qsimcirq.QSimSimulator()
In [5]: r = s.simulate(c, initial_state=0); r
Out[5]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩
In [6]: r = s.simulate(c, initial_state=r.state_vector()); r
Out[6]:
measurements: (no measurements)
output vector: (0.5+0.5j)|0⟩ + (0.5-0.5j)|1⟩
In [7]: r = s.simulate(c, initial_state=r.state_vector()); r
Out[7]:
measurements: (no measurements)
output vector: (0.146+0.354j)|0⟩ + (0.854-0.354j)|1⟩
In [8]: r = s.simulate(c, initial_state=_13.state_vector()); r
Out[8]:
measurements: (no measurements)
output vector: |1⟩95-martin-orion commented
@maffoo, could you try this with 2 and 3 qubits and report the results? qsimcirq ought to be AVX/SSE agnostic, but it's possible that certain operations in the Python-to-C++ translation are specific to AVX.
I'm specifically suspicious of NormalToInternalOrder, which mangles the qubit order to better align with AVX/SSE. If it's not actually needed for SSE, what I expect to see with 2 or 3 qubits is that initial state affects the result (unlike above), but the effect is incorrect.
maffoo commented
I see the same thing with 2 or 3 qubits:
In [2]: import cirq; import qsimcirq
In [1]: q0, q1, q2 = cirq.LineQubit.range(3)
In [3]: c2 = cirq.Circuit(cirq.X(q0)**0.25, cirq.Y(q1)**0.5); c3 = cirq.Circuit(cirq.X(q0)**0.25, cirq.Y(q1)**0.5, cirq.X(q2)**0.125)
In [4]: s = qsimcirq.QSimSimulator()
In [5]: r = s.simulate(c2, initial_state=0); r
Out[5]:
measurements: (no measurements)
output vector: (0.25+0.604j)|00⟩ + (0.25+0.604j)|01⟩ + (0.25-0.104j)|10⟩ + (0.25-0.104j)|11⟩
In [6]: r = s.simulate(c2, initial_state=r.state_vector()); r
Out[6]:
measurements: (no measurements)
output vector: (0.25+0.604j)|00⟩ + (0.25+0.604j)|01⟩ + (0.25-0.104j)|10⟩ + (0.25-0.104j)|11⟩
In [7]: r = s.simulate(c3, initial_state=0); r
Out[7]:
measurements: (no measurements)
output vector: (0.125+0.628j)|000⟩ + (0.125-0.025j)|001⟩ + (0.125+0.628j)|010⟩ + (0.125-0.025j)|011⟩ + (0.26-0.052j)|100⟩ + (-0.01-0.052j)|101⟩ + (0.26-0.052j)|110⟩ + (-0.01-0.052j)|111⟩
In [8]: r = s.simulate(c3, initial_state=r.state_vector()); r
Out[8]:
measurements: (no measurements)
output vector: (0.125+0.628j)|000⟩ + (0.125-0.025j)|001⟩ + (0.125+0.628j)|010⟩ + (0.125-0.025j)|011⟩ + (0.26-0.052j)|100⟩ + (-0.01-0.052j)|101⟩ + (0.26-0.052j)|110⟩ + (-0.01-0.052j)|111⟩maffoo commented
Looks like the sse simulator also ignores integer initial states:
In [20]: c = cirq.Circuit(cirq.X(q0)**0.25)
In [21]: r = s.simulate(c, initial_state=0); r
Out[21]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩
In [22]: r = s.simulate(c, initial_state=1); r
Out[22]:
measurements: (no measurements)
output vector: (0.854+0.354j)|0⟩ + (0.146-0.354j)|1⟩maffoo commented
Ugh, I was inadvertently using an old version of qsim (0.6.0). Tried again with 0.9.2 and setting initial_state works as expected. Sorry for the false alarm! :-)