diff --git a/basis/compiler/cfg/intrinsics/intrinsics.factor b/basis/compiler/cfg/intrinsics/intrinsics.factor index bae456ffd5..b320489080 100644 --- a/basis/compiler/cfg/intrinsics/intrinsics.factor +++ b/basis/compiler/cfg/intrinsics/intrinsics.factor @@ -177,7 +177,12 @@ IN: compiler.cfg.intrinsics { math.vectors.simd.intrinsics:(simd-vor) [ [ ^^or-vector ] emit-binary-vector-op ] } { math.vectors.simd.intrinsics:(simd-vxor) [ [ ^^xor-vector ] emit-binary-vector-op ] } { math.vectors.simd.intrinsics:(simd-vnot) [ [ ^^not-vector ] emit-unary-vector-op ] } + { math.vectors.simd.intrinsics:(simd-v<=) [ [ cc<= ^^compare-vector ] emit-binary-vector-op ] } + { math.vectors.simd.intrinsics:(simd-v<) [ [ cc< ^^compare-vector ] emit-binary-vector-op ] } { math.vectors.simd.intrinsics:(simd-v=) [ [ cc= ^^compare-vector ] emit-binary-vector-op ] } + { math.vectors.simd.intrinsics:(simd-v>) [ [ cc> ^^compare-vector ] emit-binary-vector-op ] } + { math.vectors.simd.intrinsics:(simd-v>=) [ [ cc>= ^^compare-vector ] emit-binary-vector-op ] } + { math.vectors.simd.intrinsics:(simd-vunordered?) [ [ cc/<>= ^^compare-vector ] emit-binary-vector-op ] } { math.vectors.simd.intrinsics:(simd-vany?) [ [ vcc-any ^^test-vector ] emit-unary-vector-op ] } { math.vectors.simd.intrinsics:(simd-vall?) [ [ vcc-all ^^test-vector ] emit-unary-vector-op ] } { math.vectors.simd.intrinsics:(simd-vnone?) [ [ vcc-none ^^test-vector ] emit-unary-vector-op ] } diff --git a/basis/compiler/tree/propagation/simd/simd.factor b/basis/compiler/tree/propagation/simd/simd.factor index f4495701e8..06d96ef28e 100644 --- a/basis/compiler/tree/propagation/simd/simd.factor +++ b/basis/compiler/tree/propagation/simd/simd.factor @@ -31,7 +31,12 @@ IN: compiler.tree.propagation.simd (simd-hlshift) (simd-hrshift) (simd-vshuffle) + (simd-v<=) + (simd-v<) (simd-v=) + (simd-v>) + (simd-v>=) + (simd-vunordered?) (simd-with) (simd-gather-2) (simd-gather-4) diff --git a/basis/cpu/architecture/architecture.factor b/basis/cpu/architecture/architecture.factor index 30ec0c8968..1d683ffbe3 100644 --- a/basis/cpu/architecture/architecture.factor +++ b/basis/cpu/architecture/architecture.factor @@ -256,7 +256,7 @@ HOOK: %fill-vector-reps cpu ( -- reps ) HOOK: %gather-vector-2-reps cpu ( -- reps ) HOOK: %gather-vector-4-reps cpu ( -- reps ) HOOK: %shuffle-vector-reps cpu ( -- reps ) -HOOK: %compare-vector-reps cpu ( -- reps ) +HOOK: %compare-vector-reps cpu ( cc -- reps ) HOOK: %test-vector-reps cpu ( -- reps ) HOOK: %add-vector-reps cpu ( -- reps ) HOOK: %saturated-add-vector-reps cpu ( -- reps ) diff --git a/basis/cpu/ppc/ppc.factor b/basis/cpu/ppc/ppc.factor index f41a616fcd..2f00ee0591 100644 --- a/basis/cpu/ppc/ppc.factor +++ b/basis/cpu/ppc/ppc.factor @@ -262,7 +262,7 @@ M: ppc %fill-vector-reps { } ; M: ppc %gather-vector-2-reps { } ; M: ppc %gather-vector-4-reps { } ; M: ppc %shuffle-vector-reps { } ; -M: ppc %compare-vector-reps { } ; +M: ppc %compare-vector-reps drop { } ; M: ppc %test-vector-reps { } ; M: ppc %add-vector-reps { } ; M: ppc %saturated-add-vector-reps { } ; diff --git a/basis/cpu/x86/x86.factor b/basis/cpu/x86/x86.factor index 631a428b84..f7d1aabfdd 100644 --- a/basis/cpu/x86/x86.factor +++ b/basis/cpu/x86/x86.factor @@ -770,12 +770,30 @@ M: x86 %compare-vector ( dst src1 src2 temp rep cc -- ) [ %compare-float-vector ] [ %compare-int-vector ] if ; -M: x86 %compare-vector-reps +: %compare-vector-eq-reps ( -- reps ) { { sse? { float-4-rep } } { sse2? { double-2-rep char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep } } { sse4.1? { longlong-2-rep ulonglong-2-rep } } } available-reps ; +: %compare-vector-unord-reps ( -- reps ) + { + { sse? { float-4-rep } } + { sse2? { double-2-rep } } + } available-reps ; +: %compare-vector-ord-reps ( -- reps ) + { + { sse? { float-4-rep } } + { sse2? { double-2-rep char-16-rep short-8-rep int-4-rep } } + { sse4.1? { longlong-2-rep } } + } available-reps ; + +M: x86 %compare-vector-reps + { + { [ dup { cc= cc/= } memq? ] [ drop %compare-vector-eq-reps ] } + { [ dup { cc<>= cc/<>= } memq? ] [ drop %compare-vector-unord-reps ] } + [ drop %compare-vector-ord-reps ] + } cond ; :: %test-vector-mask ( dst temp mask vcc -- ) vcc { diff --git a/basis/math/vectors/simd/intrinsics/intrinsics.factor b/basis/math/vectors/simd/intrinsics/intrinsics.factor index 0ca342fde2..0b6b897c4b 100644 --- a/basis/math/vectors/simd/intrinsics/intrinsics.factor +++ b/basis/math/vectors/simd/intrinsics/intrinsics.factor @@ -1,9 +1,9 @@ ! Copyright (C) 2009 Slava Pestov. ! See http://factorcode.org/license.txt for BSD license. USING: alien alien.c-types alien.data assocs combinators -cpu.architecture fry generalizations kernel libc macros math -sequences effects accessors namespaces lexer parser vocabs.parser -words arrays math.vectors ; +cpu.architecture compiler.cfg.comparisons fry generalizations +kernel libc macros math sequences effects accessors namespaces +lexer parser vocabs.parser words arrays math.vectors ; IN: math.vectors.simd.intrinsics ERROR: bad-simd-call ; @@ -55,7 +55,12 @@ SIMD-OP: vrshift SIMD-OP: hlshift SIMD-OP: hrshift SIMD-OP: vshuffle +SIMD-OP: v<= +SIMD-OP: v< SIMD-OP: v= +SIMD-OP: v> +SIMD-OP: v>= +SIMD-OP: vunordered? SIMD-OP: vany? SIMD-OP: vall? SIMD-OP: vnone? @@ -113,39 +118,44 @@ GENERIC# supported-simd-op? 1 ( rep intrinsic -- ? ) M: vector-rep supported-simd-op? { - { \ (simd-v+) [ %add-vector-reps ] } - { \ (simd-vs+) [ %saturated-add-vector-reps ] } - { \ (simd-v+-) [ %add-sub-vector-reps ] } - { \ (simd-v-) [ %sub-vector-reps ] } - { \ (simd-vs-) [ %saturated-sub-vector-reps ] } - { \ (simd-v*) [ %mul-vector-reps ] } - { \ (simd-vs*) [ %saturated-mul-vector-reps ] } - { \ (simd-v/) [ %div-vector-reps ] } - { \ (simd-vmin) [ %min-vector-reps ] } - { \ (simd-vmax) [ %max-vector-reps ] } - { \ (simd-v.) [ %dot-vector-reps ] } - { \ (simd-vsqrt) [ %sqrt-vector-reps ] } - { \ (simd-sum) [ %horizontal-add-vector-reps ] } - { \ (simd-vabs) [ %abs-vector-reps ] } - { \ (simd-vbitand) [ %and-vector-reps ] } - { \ (simd-vbitandn) [ %andn-vector-reps ] } - { \ (simd-vbitor) [ %or-vector-reps ] } - { \ (simd-vbitxor) [ %xor-vector-reps ] } - { \ (simd-vbitnot) [ %not-vector-reps ] } - { \ (simd-vand) [ %and-vector-reps ] } - { \ (simd-vandn) [ %andn-vector-reps ] } - { \ (simd-vor) [ %or-vector-reps ] } - { \ (simd-vxor) [ %xor-vector-reps ] } - { \ (simd-vnot) [ %not-vector-reps ] } - { \ (simd-vlshift) [ %shl-vector-reps ] } - { \ (simd-vrshift) [ %shr-vector-reps ] } - { \ (simd-hlshift) [ %horizontal-shl-vector-reps ] } - { \ (simd-hrshift) [ %horizontal-shr-vector-reps ] } - { \ (simd-vshuffle) [ %shuffle-vector-reps ] } - { \ (simd-v=) [ %compare-vector-reps ] } - { \ (simd-gather-2) [ %gather-vector-2-reps ] } - { \ (simd-gather-4) [ %gather-vector-4-reps ] } - { \ (simd-vany?) [ %test-vector-reps ] } - { \ (simd-vall?) [ %test-vector-reps ] } - { \ (simd-vnone?) [ %test-vector-reps ] } + { \ (simd-v+) [ %add-vector-reps ] } + { \ (simd-vs+) [ %saturated-add-vector-reps ] } + { \ (simd-v+-) [ %add-sub-vector-reps ] } + { \ (simd-v-) [ %sub-vector-reps ] } + { \ (simd-vs-) [ %saturated-sub-vector-reps ] } + { \ (simd-v*) [ %mul-vector-reps ] } + { \ (simd-vs*) [ %saturated-mul-vector-reps ] } + { \ (simd-v/) [ %div-vector-reps ] } + { \ (simd-vmin) [ %min-vector-reps ] } + { \ (simd-vmax) [ %max-vector-reps ] } + { \ (simd-v.) [ %dot-vector-reps ] } + { \ (simd-vsqrt) [ %sqrt-vector-reps ] } + { \ (simd-sum) [ %horizontal-add-vector-reps ] } + { \ (simd-vabs) [ %abs-vector-reps ] } + { \ (simd-vbitand) [ %and-vector-reps ] } + { \ (simd-vbitandn) [ %andn-vector-reps ] } + { \ (simd-vbitor) [ %or-vector-reps ] } + { \ (simd-vbitxor) [ %xor-vector-reps ] } + { \ (simd-vbitnot) [ %not-vector-reps ] } + { \ (simd-vand) [ %and-vector-reps ] } + { \ (simd-vandn) [ %andn-vector-reps ] } + { \ (simd-vor) [ %or-vector-reps ] } + { \ (simd-vxor) [ %xor-vector-reps ] } + { \ (simd-vnot) [ %not-vector-reps ] } + { \ (simd-vlshift) [ %shl-vector-reps ] } + { \ (simd-vrshift) [ %shr-vector-reps ] } + { \ (simd-hlshift) [ %horizontal-shl-vector-reps ] } + { \ (simd-hrshift) [ %horizontal-shr-vector-reps ] } + { \ (simd-vshuffle) [ %shuffle-vector-reps ] } + { \ (simd-v<=) [ cc<= %compare-vector-reps ] } + { \ (simd-v<) [ cc< %compare-vector-reps ] } + { \ (simd-v=) [ cc= %compare-vector-reps ] } + { \ (simd-v>) [ cc> %compare-vector-reps ] } + { \ (simd-v>=) [ cc>= %compare-vector-reps ] } + { \ (simd-vunordered?) [ cc/<>= %compare-vector-reps ] } + { \ (simd-gather-2) [ %gather-vector-2-reps ] } + { \ (simd-gather-4) [ %gather-vector-4-reps ] } + { \ (simd-vany?) [ %test-vector-reps ] } + { \ (simd-vall?) [ %test-vector-reps ] } + { \ (simd-vnone?) [ %test-vector-reps ] } } case member? ; diff --git a/basis/math/vectors/specialization/specialization.factor b/basis/math/vectors/specialization/specialization.factor index 6bec603ff7..217849ab3d 100644 --- a/basis/math/vectors/specialization/specialization.factor +++ b/basis/math/vectors/specialization/specialization.factor @@ -98,7 +98,12 @@ H{ { hrshift { +vector+ +literal+ -> +vector+ } } { vshuffle { +vector+ +literal+ -> +vector+ } } { vbroadcast { +vector+ +literal+ -> +vector+ } } + { v<= { +vector+ +vector+ -> +vector+ } } + { v< { +vector+ +vector+ -> +vector+ } } { v= { +vector+ +vector+ -> +vector+ } } + { v> { +vector+ +vector+ -> +vector+ } } + { v>= { +vector+ +vector+ -> +vector+ } } + { vunordered? { +vector+ +vector+ -> +vector+ } } { vany? { +vector+ -> +scalar+ } } { vall? { +vector+ -> +scalar+ } } { vnone? { +vector+ -> +scalar+ } }