steve
/
factor
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

don't use intrinsics for simd vector element operations when the component size doesn't fit in a fixnum

db4
Joe Groff 2009-12-03 12:46:56 -08:00
parent 7c8d755c45
commit 63b89c65f9
2 changed files with 13 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
basis/compiler/cfg/intrinsics/simd/simd.factor
View File

@ -1,7 +1,7 @@
! Copyright (C) 2009 Slava Pestov, Joe Groff. ! Copyright (C) 2009 Slava Pestov, Joe Groff.
! See http://factorcode.org/license.txt for BSD license. ! See http://factorcode.org/license.txt for BSD license.
USING: accessors alien alien.c-types byte-arrays fry USING: accessors alien alien.c-types byte-arrays fry
classes.algebra cpu.architecture kernel math sequences classes.algebra cpu.architecture kernel layouts math sequences
math.vectors math.vectors.simd.intrinsics math.vectors math.vectors.simd.intrinsics
macros generalizations combinators combinators.short-circuit macros generalizations combinators combinators.short-circuit
arrays locals compiler.tree.propagation.info arrays locals compiler.tree.propagation.info
@ -152,6 +152,9 @@ IN: compiler.cfg.intrinsics.simd
] } ] }
} v-vector-op ; } v-vector-op ;
PREDICATE: fixnum-vector-rep < int-vector-rep
rep-component-type heap-size cell < ;
: ^(sum-vector-2) ( src rep -- dst ) : ^(sum-vector-2) ( src rep -- dst )
{ {
[ dupd ^^horizontal-add-vector ] [ dupd ^^horizontal-add-vector ]
@ -249,7 +252,7 @@ IN: compiler.cfg.intrinsics.simd
: ^sum-vector ( src rep -- dst ) : ^sum-vector ( src rep -- dst )
{ {
{ float-vector-rep [ ^(sum-vector) ] } { float-vector-rep [ ^(sum-vector) ] }
{ int-vector-rep [| src rep | { fixnum-vector-rep [| src rep |
src rep ^unpack-vector-head :> head src rep ^unpack-vector-head :> head
src rep ^unpack-vector-tail :> tail src rep ^unpack-vector-tail :> tail
rep widen-vector-rep :> wide-rep rep widen-vector-rep :> wide-rep
@ -526,22 +529,26 @@ IN: compiler.cfg.intrinsics.simd
: emit-simd-with ( node -- ) : emit-simd-with ( node -- )
{ {
[ ^with-vector ] { fixnum-vector-rep [ ^with-vector ] }
{ float-vector-rep [ ^with-vector ] }
} emit-v-vector-op ; } emit-v-vector-op ;
: emit-simd-gather-2 ( node -- ) : emit-simd-gather-2 ( node -- )
{ {
[ ^^gather-vector-2 ] { fixnum-vector-rep [ ^^gather-vector-2 ] }
{ float-vector-rep [ ^^gather-vector-2 ] }
} emit-vv-vector-op ; } emit-vv-vector-op ;
: emit-simd-gather-4 ( node -- ) : emit-simd-gather-4 ( node -- )
{ {
[ ^^gather-vector-4 ] { fixnum-vector-rep [ ^^gather-vector-4 ] }
{ float-vector-rep [ ^^gather-vector-4 ] }
} emit-vvvv-vector-op ; } emit-vvvv-vector-op ;
: emit-simd-select ( node -- ) : emit-simd-select ( node -- )
{ {
[ ^select-vector ] { fixnum-vector-rep [ ^select-vector ] }
{ float-vector-rep [ ^select-vector ] }
} [ integer? ] emit-vl-vector-op ; } [ integer? ] emit-vl-vector-op ;
: emit-alien-vector ( node -- ) : emit-alien-vector ( node -- )

7
basis/cpu/x86/x86.factor
View File

@ -1323,15 +1323,8 @@ M: x86 %shr-vector-imm-reps %shr-vector-reps ;
M: x86 %integer>scalar drop MOVD ; M: x86 %integer>scalar drop MOVD ;
! XXX the longlong versions won't work on x86.32
M:: x86 %scalar>integer ( dst src rep -- ) M:: x86 %scalar>integer ( dst src rep -- )
rep { rep {
{ longlong-scalar-rep [
dst src MOVD
] }
{ ulonglong-scalar-rep [
dst src MOVD
] }
{ int-scalar-rep [ { int-scalar-rep [
dst 32-bit-version-of src MOVD dst 32-bit-version-of src MOVD
dst dst 32-bit-version-of dst dst 32-bit-version-of