Accelerate APPLY-QUBIT-PERMUTATION. #179
Conversation
jmbr
force-pushed
the
enhancement/accelerate-permutations
branch
3 times, most recently
from
0197001
to
6b4158e
Compare
jmbr
force-pushed
the
enhancement/accelerate-permutations
branch
from
6b4158e
to
6afe4c8
Compare
jmbr
force-pushed
the
enhancement/accelerate-permutations
branch
from
0842977
to
00c4b62
Compare
jmbr
force-pushed
the
enhancement/accelerate-permutations
branch
from
57eba3b
to
f2619b8
Compare
| (declare (type qubit-index tau)) | ||
|
|
||
| ;; Swap bits 0 and TAU in ADDRESS. | ||
| (let ((x (logxor (logand address 1) (logand (ash address (- tau)) 1)))) |
There was a problem hiding this comment.
how about (rotatef (ldb (byte 1 ...) address) (ldb (byte 1 ...) address))?
(setf ldb) is a valid function
There was a problem hiding this comment.
That's exactly how it was before but the above produces tighter assembly code.
There was a problem hiding this comment.
Ok, can you make a comment with #+#:slightly-slower (rotatef ...) or so just to demo something equivalent and a lot more readable?
| (values qvm:amplitude-address)) | ||
|
|
||
| ;; Swap bits 0 and TAU in ADDRESS. | ||
| (let ((x (logxor (logand ,address 1) (logand (ash ,address ,minus-tau) 1)))) |
There was a problem hiding this comment.
same here. Perhaps factor it out into an inlined function
| ;;; more than three times faster than the equivalent application of a | ||
| ;;; PERMUTATION-GENERAL object). | ||
|
|
||
| (defconstant +qubit-index-length+ (ceiling (log qvm::+max-nat-tuple-cardinality+ 2)) |
There was a problem hiding this comment.
use integer-length appropriately
| (defconstant +qubit-index-length+ (ceiling (log qvm::+max-nat-tuple-cardinality+ 2)) | ||
| "Number of bits required to represent a qubit index.") | ||
|
|
||
| (defconstant +max-number-of-transpositions+ (* 2 #.qvm::+max-nat-tuple-cardinality+) |
| "Upper bound on the number of transpositions defining an arbitrary permutation.") | ||
|
|
||
| (deftype qubit-index () | ||
| '(unsigned-byte #.+qubit-index-length+)) |
There was a problem hiding this comment.
quasiquoting looks a little cleaner
|
|
||
| (deftype transposition () | ||
| '(or null (cons alexandria:non-negative-fixnum | ||
| alexandria:non-negative-fixnum))) | ||
| '(or null (cons qubit-index qubit-index))) |
There was a problem hiding this comment.
how can a null transposition show up?
| (transpositions (permutation-transpositions permutation))) | ||
| (cond | ||
| ((and (null domain) (null codomain)) nil) | ||
| ((and (= 1 (length domain)) |
| (the qvm:amplitude-address (first codomain))))) | ||
| (make-instance 'permutation-transposition :tau max-index :size (1+ max-index))) | ||
| ((and (= 2 (length domain)) | ||
| (null (set-difference domain codomain)) |
There was a problem hiding this comment.
do you want difference or intersection?
| (loop :for a :of-type qubit-index :in domain | ||
| :for b :of-type qubit-index :across codomain | ||
| :unless (= a b) :do | ||
| (pushnew (cons a b) transpositions :test #'equal))) |
There was a problem hiding this comment.
I would write a transposition= function and use it here, even if it's as simple as
(defun transposition= (a b)
(and (= (car a) (car b)) (= (cdr a) (cdr b))))
| (declare (type qubit-index tau)) | ||
|
|
||
| ;; Swap bits 0 and TAU in ADDRESS. | ||
| (let ((x (logxor (logand address 1) (logand (ash address (- tau)) 1)))) |
There was a problem hiding this comment.
Ok, can you make a comment with #+#:slightly-slower (rotatef ...) or so just to demo something equivalent and a lot more readable?
| (values qvm:amplitude-address)) | ||
|
|
||
| ;; Swap bits 0 and TAU in ADDRESS. | ||
| (let ((x (logxor (logand ,address 1) (logand (ash ,address ,minus-tau) 1)))) |
There was a problem hiding this comment.
same here. Perhaps factor it out into an inlined function
| (byte ,num-bits 0) | ||
| ,address))))) | ||
|
|
||
| (defun compile-qubit-permutation (permutation) |
Part of #177.