decompose %unpack-vector-head/tail into %compare-vector/%merge-vector-head/tail or %tail>head-vector/%unpack-vector-head insns when there isn't an actual unpack insn; get rid of fake x86 implementations

basis/compiler/cfg/instructions/instructions.factor

@@ -280,6 +280,11 @@ def: dst
 use: src
 literal: shuffle rep ;
 
+PURE-INSN: ##tail>head-vector
+def: dst
+use: src
+literal: rep ;
+
 PURE-INSN: ##merge-vector-head
 def: dst
 use: src1 src2
@@ -303,13 +308,11 @@ literal: rep ;
 PURE-INSN: ##unpack-vector-head
 def: dst
 use: src
-temp: temp
 literal: rep ;
 
 PURE-INSN: ##unpack-vector-tail
 def: dst
 use: src
-temp: temp
 literal: rep ;
 
 PURE-INSN: ##integer>float-vector

basis/compiler/cfg/intrinsics/intrinsics.factor

@@ -200,8 +200,8 @@ IN: compiler.cfg.intrinsics
         { math.vectors.simd.intrinsics:(simd-(v>integer)) [ [ ^^float>integer-vector ] emit-unary-vector-op ] }
         { math.vectors.simd.intrinsics:(simd-(vpack-signed)) [ [ ^^signed-pack-vector ] emit-binary-vector-op ] }
         { math.vectors.simd.intrinsics:(simd-(vpack-unsigned)) [ [ ^^unsigned-pack-vector ] emit-binary-vector-op ] }
-        { math.vectors.simd.intrinsics:(simd-(vunpack-head)) [ [ ^^unpack-vector-head ] emit-unary-vector-op ] }
+        { math.vectors.simd.intrinsics:(simd-(vunpack-head)) [ [ generate-unpack-vector-head ] emit-unary-vector-op ] }
-        { math.vectors.simd.intrinsics:(simd-(vunpack-tail)) [ [ ^^unpack-vector-tail ] emit-unary-vector-op ] }
+        { math.vectors.simd.intrinsics:(simd-(vunpack-tail)) [ [ generate-unpack-vector-tail ] emit-unary-vector-op ] }
         { math.vectors.simd.intrinsics:(simd-select) [ emit-select-vector ] }
         { math.vectors.simd.intrinsics:(simd-sum) [ [ ^^horizontal-add-vector ] emit-unary-vector-op ] }
         { math.vectors.simd.intrinsics:alien-vector [ emit-alien-vector ] }

basis/compiler/cfg/intrinsics/simd/simd.factor

@@ -2,8 +2,9 @@
 ! See http://factorcode.org/license.txt for BSD license.
 USING: accessors byte-arrays fry cpu.architecture kernel math
 sequences math.vectors.simd.intrinsics macros generalizations
-combinators combinators.short-circuit arrays
+combinators combinators.short-circuit arrays locals
 compiler.tree.propagation.info compiler.cfg.builder.blocks
+compiler.cfg.comparisons
 compiler.cfg.stacks compiler.cfg.stacks.local compiler.cfg.hats
 compiler.cfg.instructions compiler.cfg.registers
 compiler.cfg.intrinsics.alien ;
@@ -121,3 +122,36 @@ MACRO: if-literals-match ( quots -- )
     [ ^^not-vector ]
     [ [ ^^fill-vector ] [ ^^xor-vector ] bi ] if ;
 
+:: generate-unpack-vector-head ( src rep -- dst )
+    {
+        {
+            [ rep %unpack-vector-head-reps member? ]
+            [ src rep ^^unpack-vector-head ]
+        }
+        [
+            rep ^^zero-vector :> zero
+            zero src rep cc> ^^compare-vector :> sign
+            src sign rep ^^merge-vector-head
+        ] 
+    } cond ;
+
+:: generate-unpack-vector-tail ( src rep -- dst )
+    {
+        {
+            [ rep %unpack-vector-tail-reps member? ]
+            [ src rep ^^unpack-vector-tail ]
+        }
+        {
+            [ rep %unpack-vector-head-reps member? ]
+            [
+                src rep ^^tail>head-vector :> tail
+                tail rep ^^unpack-vector-head
+            ]
+        }
+        [
+            rep ^^zero-vector :> zero
+            zero src rep cc> ^^compare-vector :> sign
+            src sign rep ^^merge-vector-tail
+        ] 
+    } cond ;
+

basis/compiler/codegen/codegen.factor

@@ -164,6 +164,7 @@ CODEGEN: ##fill-vector %fill-vector
 CODEGEN: ##gather-vector-2 %gather-vector-2
 CODEGEN: ##gather-vector-4 %gather-vector-4
 CODEGEN: ##shuffle-vector %shuffle-vector
+CODEGEN: ##tail>head-vector %tail>head-vector
 CODEGEN: ##merge-vector-head %merge-vector-head
 CODEGEN: ##merge-vector-tail %merge-vector-tail
 CODEGEN: ##signed-pack-vector %signed-pack-vector

basis/cpu/architecture/architecture.factor

@@ -230,12 +230,13 @@ HOOK: %fill-vector cpu ( dst rep -- )
 HOOK: %gather-vector-2 cpu ( dst src1 src2 rep -- )
 HOOK: %gather-vector-4 cpu ( dst src1 src2 src3 src4 rep -- )
 HOOK: %shuffle-vector cpu ( dst src shuffle rep -- )
+HOOK: %tail>head-vector cpu ( dst src rep -- )
 HOOK: %merge-vector-head cpu ( dst src1 src2 rep -- )
 HOOK: %merge-vector-tail cpu ( dst src1 src2 rep -- )
 HOOK: %signed-pack-vector cpu ( dst src1 src2 rep -- )
 HOOK: %unsigned-pack-vector cpu ( dst src1 src2 rep -- )
-HOOK: %unpack-vector-head cpu ( dst src temp rep -- )
+HOOK: %unpack-vector-head cpu ( dst src rep -- )
-HOOK: %unpack-vector-tail cpu ( dst src temp rep -- )
+HOOK: %unpack-vector-tail cpu ( dst src rep -- )
 HOOK: %integer>float-vector cpu ( dst src rep -- )
 HOOK: %float>integer-vector cpu ( dst src rep -- )
 HOOK: %compare-vector cpu ( dst src1 src2 temp rep cc -- )
@@ -279,7 +280,8 @@ HOOK: %shuffle-vector-reps cpu ( -- reps )
 HOOK: %merge-vector-reps cpu ( -- reps )
 HOOK: %signed-pack-vector-reps cpu ( -- reps )
 HOOK: %unsigned-pack-vector-reps cpu ( -- reps )
-HOOK: %unpack-vector-reps cpu ( -- reps )
+HOOK: %unpack-vector-head-reps cpu ( -- reps )
+HOOK: %unpack-vector-tail-reps cpu ( -- reps )
 HOOK: %integer>float-vector-reps cpu ( -- reps )
 HOOK: %float>integer-vector-reps cpu ( -- reps )
 HOOK: %compare-vector-reps cpu ( cc -- reps )

basis/cpu/x86/x86.factor

@@ -774,45 +774,32 @@ M: x86 %unsigned-pack-vector-reps
         { sse4.1? { int-4-rep } }
     } available-reps ;
 
-:: %sign-extension-vector ( dst src rep -- )
+M: x86 %tail>head-vector ( dst src rep -- )
-    dst rep %zero-vector
+    dup {
-    dst src rep {
+        { float-4-rep [ drop MOVHLPS ] }
-        { char-16-rep    [ PCMPGTB ] }
+        { double-2-rep [ [ %copy ] [ drop UNPCKHPD ] 3bi ] }
-        { short-8-rep    [ PCMPGTW ] }
+        [ drop [ %copy ] [ drop PUNPCKHQDQ ] 3bi ]
-        { int-4-rep      [ PCMPGTD ] }
-        { longlong-2-rep [ PCMPGTQ ] }
     } case ;
 
-:: (%unpack-vector-signs) ( dst src rep -- )
+M: x86 %unpack-vector-head ( dst src rep -- )
-    dst rep signed-int-vector-rep?
-    [ src rep %sign-extension-vector ]
-    [ rep %zero-vector ] if ;
-
-M:: x86 %unpack-vector-head ( dst src temp rep -- )
-    temp src rep (%unpack-vector-signs)
-    dst src rep %copy
-    dst temp rep unsign-rep {
-        { char-16-rep    [ PUNPCKLBW ] }
-        { short-8-rep    [ PUNPCKLWD ] }
-        { int-4-rep      [ PUNPCKLDQ ] }
-        { longlong-2-rep [ PUNPCKLQDQ ] }
-    } case ;
-
-M:: x86 %unpack-vector-tail ( dst src temp rep -- )
-    temp src rep (%unpack-vector-signs)
-    dst src rep %copy
-    dst temp rep unsign-rep {
-        { char-16-rep    [ PUNPCKHBW ] }
-        { short-8-rep    [ PUNPCKHWD ] }
-        { int-4-rep      [ PUNPCKHDQ ] }
-        { longlong-2-rep [ PUNPCKHQDQ ] }
-    } case ;
-
-M: x86 %unpack-vector-reps ( -- reps )
     {
-        { sse2? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep longlong-2-rep ulonglong-2-rep } }
+        { char-16-rep  [ PMOVSXBW ] }
+        { uchar-16-rep [ PMOVZXBW ] }
+        { short-8-rep  [ PMOVSXWD ] }
+        { ushort-8-rep [ PMOVZXWD ] }
+        { int-4-rep    [ PMOVSXDQ ] }
+        { uint-4-rep   [ PMOVZXDQ ] }
+        { float-4-rep  [ CVTPS2PD ] }
+    } case ;
+
+M: x86 %unpack-vector-head-reps ( -- reps )
+    {
+        { sse2? { float-4-rep } }
+        { sse4.1? { char-16-rep uchar-16-rep short-8-rep ushort-8-rep int-4-rep uint-4-rep } }
     } available-reps ;
 
+M: x86 %unpack-vector-tail-reps ( -- reps ) { } ;
+
 M: x86 %integer>float-vector ( dst src rep -- )
     {
         { int-4-rep [ CVTDQ2PS ] }

basis/math/vectors/simd/intrinsics/intrinsics.factor

@@ -4,7 +4,7 @@ USING: alien alien.c-types alien.data assocs combinators
 cpu.architecture compiler.cfg.comparisons fry generalizations
 kernel libc macros math
 math.vectors.conversion.backend
-sequences effects accessors namespaces
+sequences sets effects accessors namespaces
 lexer parser vocabs.parser words arrays math.vectors ;
 IN: math.vectors.simd.intrinsics
 
@@ -137,6 +137,10 @@ MACRO: (simd-boa) ( rep -- quot )
 
 GENERIC# supported-simd-op? 1 ( rep intrinsic -- ? )
 
+: (%unpack-reps) ( -- reps )
+    %merge-vector-reps [ int-vector-rep? ] filter
+    %unpack-vector-head-reps union ;
+
 M: vector-rep supported-simd-op?
     {
         { \ (simd-v+)            [ %add-vector-reps            ] }
@@ -174,8 +178,8 @@ M: vector-rep supported-simd-op?
         { \ (simd-(v>integer))      [ %float>integer-vector-reps ] }
         { \ (simd-(vpack-signed))   [ %signed-pack-vector-reps   ] }
         { \ (simd-(vpack-unsigned)) [ %unsigned-pack-vector-reps ] }
-        { \ (simd-(vunpack-head))   [ %unpack-vector-reps        ] }
+        { \ (simd-(vunpack-head))   [ (%unpack-reps)             ] }
-        { \ (simd-(vunpack-tail))   [ %unpack-vector-reps        ] }
+        { \ (simd-(vunpack-tail))   [ (%unpack-reps)             ] }
         { \ (simd-v<=)           [ cc<= %compare-vector-reps   ] }
         { \ (simd-v<)            [ cc< %compare-vector-reps    ] }
         { \ (simd-v=)            [ cc= %compare-vector-reps    ] }