suzanne.soy/racket
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix optimization of unary complex - and /.

Browse Source
This commit is contained in:
Vincent St-Amour 2013-02-14 17:19:58 -05:00
parent c6029cacf8
commit 60959467cf
3 changed files with 187 additions and 65 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

147
collects/tests/typed-racket/optimizer/tests/pr13468.rkt Normal file
View File

@ -0,0 +1,147 @@
#;
(
TR opt: pr13468.rkt 86:13 6.0+2.3i -- unboxed literal
TR opt: pr13468.rkt 86:13 6.0+2.3i -- unboxed literal
TR opt: pr13468.rkt 86:5 (- (ann 6.0+2.3i Float-Complex)) -- unboxed unary float complex
TR opt: pr13468.rkt 87:13 6.0+2.3i -- unboxed literal
TR opt: pr13468.rkt 87:13 6.0+2.3i -- unboxed literal
TR opt: pr13468.rkt 87:5 (/ (ann 6.0+2.3i Float-Complex)) -- unboxed unary float complex
TR opt: pr13468.rkt 88:13 0.0+0.0i -- unboxed literal
TR opt: pr13468.rkt 88:42 1 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 88:5 (/ (ann 0.0+0.0i Float-Complex) (ann 1 Number)) -- unboxed binary float complex
TR missed opt: pr13468.rkt 91:5 (expt (ann -5.0 Flonum) (ann 2.0 Flonum)) -- unexpected complex type
TR opt: pr13468.rkt 112:21 0.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 112:5 (magnitude (ann 0.0 Flonum-Zero)) -- unboxed unary float complex
TR opt: pr13468.rkt 113:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 113:5 (magnitude (ann 6.0 Positive-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 114:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 114:5 (magnitude (ann 6.0 Nonnegative-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 115:21 -6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 115:5 (magnitude (ann -6.0 Nonpositive-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 116:21 -6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 116:5 (magnitude (ann -6.0 Negative-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 117:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 117:5 (magnitude (ann 6.0 Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 138:21 0.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 138:5 (real-part (ann 0.0 Flonum-Zero)) -- unboxed unary float complex
TR opt: pr13468.rkt 139:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 139:5 (real-part (ann 6.0 Positive-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 140:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 140:5 (real-part (ann 6.0 Nonnegative-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 141:21 -6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 141:5 (real-part (ann -6.0 Nonpositive-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 142:21 -6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 142:5 (real-part (ann -6.0 Negative-Flonum)) -- unboxed unary float complex
TR opt: pr13468.rkt 143:21 6.0 -- float-arg-expr in complex ops
TR opt: pr13468.rkt 143:5 (real-part (ann 6.0 Flonum)) -- unboxed unary float complex
-6.0-2.3i
0.1453136352627755-0.05570356018406394i
0.0+0.0i
25.0
0
1
6
6
6
6
6
6
6
6
6
6
6
6
0.0
6.0
6.0
6.0
6.0
6.0
0
1
6
6
6
6
-6
-6
6
6
6
-6
-6
6
0.0
6.0
6.0
-6.0
-6.0
6.0
0)
#lang typed/racket
;; Most bothersome missing cases:
(ann (- (ann 6.0+2.3i Float-Complex)) Float-Complex)
(ann (/ (ann 6.0+2.3i Float-Complex)) Float-Complex)
(ann (/ (ann 0.0+0.0i Float-Complex) (ann 1+0i Number)) Float-Complex)
;; Would be nice to maintain flonum-ness here:
(ann (expt (ann -5.0 Flonum) (ann 2.0 Flonum)) (U Flonum Float-Complex))
;; In general, for real numbers, `magnitude' should have the same cases as `abs'
(ann (magnitude (ann 0 Zero)) Zero)
(ann (magnitude (ann 1 One)) One)
(ann (magnitude (ann 6 Byte)) Byte)
(ann (magnitude (ann 6 Index)) Index)
(ann (magnitude (ann 6 Positive-Fixnum)) Positive-Fixnum)
(ann (magnitude (ann 6 Nonnegative-Fixnum)) Nonnegative-Fixnum)
(ann (magnitude (ann -6 Nonpositive-Fixnum)) Nonnegative-Integer)
(ann (magnitude (ann -6 Negative-Fixnum)) Positive-Integer)
(ann (magnitude (ann 6 Fixnum)) Integer)
(ann (magnitude (ann 6 Positive-Integer)) Positive-Integer)
(ann (magnitude (ann 6 Nonnegative-Integer)) Nonnegative-Integer)
(ann (magnitude (ann -6 Nonpositive-Integer)) Nonnegative-Integer)
(ann (magnitude (ann -6 Negative-Integer)) Positive-Integer)
(ann (magnitude (ann 6 Integer)) Integer)
(ann (magnitude (ann 0.0 Flonum-Zero)) Flonum-Zero)
(ann (magnitude (ann 6.0 Positive-Flonum)) Positive-Flonum)
(ann (magnitude (ann 6.0 Nonnegative-Flonum)) Nonnegative-Flonum)
(ann (magnitude (ann -6.0 Nonpositive-Flonum)) Nonnegative-Flonum)
(ann (magnitude (ann -6.0 Negative-Flonum)) Positive-Flonum)
(ann (magnitude (ann 6.0 Flonum)) Flonum)
;; In general, for real numbers, `real-part' should return the same type
(ann (real-part (ann 0 Zero)) Zero)
(ann (real-part (ann 1 One)) One)
(ann (real-part (ann 6 Byte)) Byte)
(ann (real-part (ann 6 Index)) Index)
(ann (real-part (ann 6 Positive-Fixnum)) Positive-Fixnum)
(ann (real-part (ann 6 Nonnegative-Fixnum)) Nonnegative-Fixnum)
(ann (real-part (ann -6 Nonpositive-Fixnum)) Nonpositive-Fixnum)
(ann (real-part (ann -6 Negative-Fixnum)) Negative-Fixnum)
(ann (real-part (ann 6 Fixnum)) Fixnum)
(ann (real-part (ann 6 Positive-Integer)) Positive-Integer)
(ann (real-part (ann 6 Nonnegative-Integer)) Nonnegative-Integer)
(ann (real-part (ann -6 Nonpositive-Integer)) Nonpositive-Integer)
(ann (real-part (ann -6 Negative-Integer)) Negative-Integer)
(ann (real-part (ann 6 Integer)) Integer)
(ann (real-part (ann 0.0 Flonum-Zero)) Flonum-Zero)
(ann (real-part (ann 6.0 Positive-Flonum)) Positive-Flonum)
(ann (real-part (ann 6.0 Nonnegative-Flonum)) Nonnegative-Flonum)
(ann (real-part (ann -6.0 Nonpositive-Flonum)) Nonpositive-Flonum)
(ann (real-part (ann -6.0 Negative-Flonum)) Negative-Flonum)
(ann (real-part (ann 6.0 Flonum)) Flonum)
;; This one surprised me with the exactness of its return value, but it seems
;; to be true:
(ann (imag-part (ann 5.6 Real)) Zero)

65
collects/tests/typed-racket/succeed/pr13468.rkt
View File

@ -1,65 +0,0 @@
#lang typed/racket
;; Most bothersome missing cases:
(ann (- (ann 6.0+2.3i Float-Complex)) Float-Complex)
(ann (/ (ann 6.0+2.3i Float-Complex)) Float-Complex)
(ann (/ (ann 0.0+0.0i Float-Complex) (ann 1+0i Number)) Float-Complex)
;; Would be nice to maintain flonum-ness here:
(ann (expt (ann -5.0 Flonum) (ann 2.0 Flonum)) (U Flonum Float-Complex))
;; In general, for real numbers, `magnitude' should have the same cases as `abs'
(ann (magnitude (ann 0 Zero)) Zero)
(ann (magnitude (ann 1 One)) One)
(ann (magnitude (ann 6 Byte)) Byte)
(ann (magnitude (ann 6 Index)) Index)
(ann (magnitude (ann 6 Positive-Fixnum)) Positive-Fixnum)
(ann (magnitude (ann 6 Nonnegative-Fixnum)) Nonnegative-Fixnum)
(ann (magnitude (ann -6 Nonpositive-Fixnum)) Nonnegative-Integer)
(ann (magnitude (ann -6 Negative-Fixnum)) Positive-Integer)
(ann (magnitude (ann 6 Fixnum)) Integer)
(ann (magnitude (ann 6 Positive-Integer)) Positive-Integer)
(ann (magnitude (ann 6 Nonnegative-Integer)) Nonnegative-Integer)
(ann (magnitude (ann -6 Nonpositive-Integer)) Nonnegative-Integer)
(ann (magnitude (ann -6 Negative-Integer)) Positive-Integer)
(ann (magnitude (ann 6 Integer)) Integer)
(ann (magnitude (ann 0.0 Flonum-Zero)) Flonum-Zero)
(ann (magnitude (ann 6.0 Positive-Flonum)) Positive-Flonum)
(ann (magnitude (ann 6.0 Nonnegative-Flonum)) Nonnegative-Flonum)
(ann (magnitude (ann -6.0 Nonpositive-Flonum)) Nonnegative-Flonum)
(ann (magnitude (ann -6.0 Negative-Flonum)) Positive-Flonum)
(ann (magnitude (ann 6.0 Flonum)) Flonum)
;; In general, for real numbers, `real-part' should return the same type
(ann (real-part (ann 0 Zero)) Zero)
(ann (real-part (ann 1 One)) One)
(ann (real-part (ann 6 Byte)) Byte)
(ann (real-part (ann 6 Index)) Index)
(ann (real-part (ann 6 Positive-Fixnum)) Positive-Fixnum)
(ann (real-part (ann 6 Nonnegative-Fixnum)) Nonnegative-Fixnum)
(ann (real-part (ann -6 Nonpositive-Fixnum)) Nonpositive-Fixnum)
(ann (real-part (ann -6 Negative-Fixnum)) Negative-Fixnum)
(ann (real-part (ann 6 Fixnum)) Fixnum)
(ann (real-part (ann 6 Positive-Integer)) Positive-Integer)
(ann (real-part (ann 6 Nonnegative-Integer)) Nonnegative-Integer)
(ann (real-part (ann -6 Nonpositive-Integer)) Nonpositive-Integer)
(ann (real-part (ann -6 Negative-Integer)) Negative-Integer)
(ann (real-part (ann 6 Integer)) Integer)
(ann (real-part (ann 0.0 Flonum-Zero)) Flonum-Zero)
(ann (real-part (ann 6.0 Positive-Flonum)) Positive-Flonum)
(ann (real-part (ann 6.0 Nonnegative-Flonum)) Nonnegative-Flonum)
(ann (real-part (ann -6.0 Nonpositive-Flonum)) Nonpositive-Flonum)
(ann (real-part (ann -6.0 Negative-Flonum)) Negative-Flonum)
(ann (real-part (ann 6.0 Flonum)) Flonum)
;; This one surprised me with the exactness of its return value, but it seems
;; to be true:
(ann (imag-part (ann 5.6 Real)) Zero)

40
collects/typed-racket/optimizer/float-complex.rkt
View File

@ -99,6 +99,18 @@
(list
#`((real-binding) #,(skip-0s #'(c1.real-binding c2.real-binding cs.real-binding ...)))
#`((imag-binding) #,(skip-0s #'(c1.imag-binding c2.imag-binding cs.imag-binding ...)))))))))
(pattern (#%plain-app (~and op (~literal -)) c1:unboxed-float-complex-opt-expr) ; unary -
#:when (subtypeof? this-syntax -FloatComplex)
#:with real-binding (unboxed-gensym "unboxed-real-")
#:with imag-binding (unboxed-gensym "unboxed-imag-")
#:with (bindings ...)
(begin (log-optimization "unboxed unary float complex"
complex-unboxing-opt-msg
this-syntax)
(add-disappeared-use #'op)
#`(c1.bindings ...
[(real-binding) (unsafe-fl- 0.0 #,(get-part-or-0.0 #'c1.real-binding))]
[(imag-binding) (unsafe-fl- 0.0 #,(get-part-or-0.0 #'c1.imag-binding))])))
(pattern (#%plain-app (~and op (~literal *))
c1:unboxed-float-complex-opt-expr
@ -225,6 +237,34 @@
(unsafe-fl+ (unsafe-fl* #,(car e1) #,(car e1))
(unsafe-fl* #,(car e2) #,(car e2))))
res)]))))))))
(pattern (#%plain-app (~and op (~literal /)) c1:unboxed-float-complex-opt-expr) ; unary /
#:when (subtypeof? this-syntax -FloatComplex)
#:with real-binding (unboxed-gensym "unboxed-real-")
#:with imag-binding (unboxed-gensym "unboxed-imag-")
#:with (bindings ...)
(begin (log-optimization "unboxed unary float complex"
complex-unboxing-opt-msg
this-syntax)
(add-disappeared-use #'op)
;; (/ 1.0+0.0i c1)
;; = (+ (/ (+ (* 1.0 c1.real) (* 0.0 c1.imag)) (+ c1.real^2 c1.imag^2))
;; (/ (- (* 0.0 c1.real) (* 1.0 c1.imag)) (+ c1.real^2 c1.imag^2))*i)
;; = (+ (/ c1.real (+ c1.real^2 c1.imag^2))
;; (/ (- 0.0 c1.imag) (+ c1.real^2 c1.imag^2))*i)
(with-syntax ([denominator-binding (unboxed-gensym)])
#`(c1.bindings ...
[(denominator-binding)
#,(cond [(not (syntax->datum #'c1.imag-binding)) ; only real part
#'(unsafe-fl* c1.real-binding c1.real-binding)]
[(not (syntax->datum #'c1.real-binding)) ; only imag part
#'(unsafe-fl* c1.imag-binding c1.imag-binding)]
[else ; both parts
#'(unsafe-fl+ (unsafe-fl* c1.real-binding c1.real-binding)
(unsafe-fl* c1.imag-binding c1.imag-binding))])]
[(real-binding) (unsafe-fl/ #,(get-part-or-0.0 #'c1.real-binding)
denominator-binding)]
[(imag-binding) (unsafe-fl/ (unsafe-fl- 0.0 #,(get-part-or-0.0 #'c1.imag-binding))
denominator-binding)]))))
(pattern (#%plain-app (~and op (~literal conjugate)) c:unboxed-float-complex-opt-expr)
#:when (subtypeof? this-syntax -FloatComplex)