@ttsc/win32-arm64 0.33.0 → 0.34.0

package/bin/go/src/simd/archsimd/_gen/simdgen/ops/Moves/{go.yaml → go_amd64.yaml}

@@ -1,6 +1,6 @@
 !sum
 - go: SetElem
-  asm: "VPINSR[BWDQ]"
+  asm: "VPINSRB"
   in:
   - &t
     class: vreg
@@ -8,15 +8,60 @@
   - class: greg
     base: $b
     lanes: 1 # Scalar, darn it!
-  - &imm
-    class: immediate
+  - class: immediate
+    immOffset: 0
+    immMax: 15
+    name: index
+  out:
+  - *t
+- go: SetElem
+  asm: "VPINSRW"
+  in:
+  - &t
+    class: vreg
+    base: $b
+  - class: greg
+    base: $b
+    lanes: 1 # Scalar, darn it!
+  - class: immediate
+    immOffset: 0
+    immMax: 7
+    name: index
+  out:
+  - *t
+- go: SetElem
+  asm: "VPINSRD"
+  in:
+  - &t
+    class: vreg
+    base: $b
+  - class: greg
+    base: $b
+    lanes: 1 # Scalar, darn it!
+  - class: immediate
     immOffset: 0
+    immMax: 3
+    name: index
+  out:
+  - *t
+- go: SetElem
+  asm: "VPINSRQ"
+  in:
+  - &t
+    class: vreg
+    base: $b
+  - class: greg
+    base: $b
+    lanes: 1 # Scalar, darn it!
+  - class: immediate
+    immOffset: 0
+    immMax: 1
     name: index
   out:
   - *t
 
 - go: SetElem
-  asm: "VPINSR[DQ]"
+  asm: "VPINSRD"
   in:
   - &t
     class: vreg
@@ -26,33 +71,122 @@
     base: int
     OverwriteBase: float
     lanes: 1 # Scalar, darn it!
-  - &imm
-    class: immediate
+  - class: immediate
+    immOffset: 0
+    immMax: 3
+    name: index
+  out:
+  - *t
+- go: SetElem
+  asm: "VPINSRQ"
+  in:
+  - &t
+    class: vreg
+    base: int
+    OverwriteBase: float
+  - class: greg
+    base: int
+    OverwriteBase: float
+    lanes: 1 # Scalar, darn it!
+  - class: immediate
     immOffset: 0
+    immMax: 1
     name: index
   out:
   - *t
 
+- go: GetElem # VPEXTRB 128 bit
+  asm: VPEXTRB
+  in:
+  - bits: 128
+    base: $b
+    class: vreg
+    elemBits: 8
+  - class: immediate
+    immOffset: 0
+    immMax: 15
+    name: index
+  out:
+    - bits: 32
+      base: $b
+      class: greg
+      overwriteBits: 8 # The XED data specification is wrong, correct it to the right bits
+
+- go: GetElem # VPEXTRW 128 bit
+  asm: VPEXTRW
+  in:
+  - bits: 128
+    base: $b
+    class: vreg
+    elemBits: 16
+  - class: immediate
+    immOffset: 0
+    immMax: 7
+    name: index
+  out:
+    - bits: 32
+      base: $b
+      class: greg
+      overwriteBits: 16 # The XED data specification is wrong, correct it to the right bits
+
 - go: GetElem
-  asm: "VPEXTR[BWDQ]"
+  asm: "VPEXTRD"
   in:
   - class: vreg
     base: $b
     elemBits: $e
-  - *imm
+  - class: immediate
+    immOffset: 0
+    immMax: 3
+    name: index
+  out:
+  - class: greg
+    base: $b
+    bits: $e
+
+- go: GetElem
+  asm: "VPEXTRQ"
+  in:
+  - class: vreg
+    base: $b
+    elemBits: $e
+  - class: immediate
+    immOffset: 0
+    immMax: 1
+    name: index
   out:
   - class: greg
     base: $b
     bits: $e
 
 - go: GetElem
-  asm: "VPEXTR[DQ]"
+  asm: "VPEXTRD"
   in:
   - class: vreg
     base: int
     elemBits: $e
     OverwriteBase: float
-  - *imm
+  - class: immediate
+    immOffset: 0
+    immMax: 3
+    name: index
+  out:
+  - class: greg
+    base: int
+    bits: $e
+    OverwriteBase: float
+
+- go: GetElem
+  asm: "VPEXTRQ"
+  in:
+  - class: vreg
+    base: int
+    elemBits: $e
+    OverwriteBase: float
+  - class: immediate
+    immOffset: 0
+    immMax: 1
+    name: index
   out:
   - class: greg
     base: int
@@ -225,73 +359,16 @@
   - *f64xN
 
 - go: Permute
-  asm: "VPERMQ|VPERMPD"
-  addDoc: !string |-
-    // The low 2 bits (values 0-3) of each element of indices is used.
+  asm: "VPERM[BWDQ]|VPERMP[SD]"
   operandOrder: "21Type1"
   in:
   - &anyindices
-    go: $t
+    class: vreg
     name: indices
     overwriteBase: uint
-  - &any4
-    go: $t
-    lanes: 4
-  out:
-  - &any
-    go: $t
-
-- go: Permute
-  asm: "VPERM[WDQ]|VPERMP[SD]"
-  addDoc: !string |-
-    // The low 3 bits (values 0-7) of each element of indices is used.
-  operandOrder: "21Type1"
-  in:
-  - *anyindices
-  - &any8
-    go: $t
-    lanes: 8
-  out:
-  - *any
-
-- go: Permute
-  asm: "VPERM[BWD]|VPERMPS"
-  addDoc: !string |-
-    // The low 4 bits (values 0-15) of each element of indices is used.
-  operandOrder: "21Type1"
-  in:
-  - *anyindices
-  - &any16
-    go: $t
-    lanes: 16
-  out:
-  - *any
-
-- go: Permute
-  asm: "VPERM[BW]"
-  addDoc: !string |-
-    // The low 5 bits (values 0-31) of each element of indices is used.
-  operandOrder: "21Type1"
-  in:
-  - *anyindices
-  - &any32
-    go: $t
-    lanes: 32
-  out:
-  - *any
-
-- go: Permute
-  asm: "VPERMB"
-  addDoc: !string |-
-    // The low 6 bits (values 0-63) of each element of indices is used.
-  operandOrder: "21Type1"
-  in:
-  - *anyindices
-  - &any64
-    go: $t
-    lanes: 64
+  - go: $t
   out:
-  - *any
+  - go: $t
 
 - go: ConcatPermute
   asm: "VPERMI2[BWDQ]|VPERMI2P[SD]"
@@ -301,19 +378,19 @@
   operandOrder: "231Type1"
   in:
   - *anyindices # result in arg 0
-  - *any
-  - *any
+  - go: $t
+  - go: $t
   out:
-  - *any
+  - go: $t
 
 - go: Compress
   asm: "VPCOMPRESS[BWDQ]|VCOMPRESSP[SD]"
   in:
     # The mask in Compress is a control mask rather than a write mask, so it's not optional.
   - class: mask
-  - *any
+  - go: $t
   out:
-  - *any
+  - go: $t
 
 # For now a non-public method because
 # (1) [OverwriteClass] must be set together with [OverwriteBase]
@@ -372,11 +449,11 @@
   in:
     # The mask in Expand is a control mask rather than a write mask, so it's not optional.
   - class: mask
-  - *any
+  - go: $t
   out:
-  - *any
+  - go: $t
 
-- go: Broadcast1To2
+- go: broadcast1To2
   asm: VPBROADCASTQ
   in:
   - class: vreg
@@ -390,7 +467,7 @@
     base: $b
 
 # weirdly, this one case on AVX2 is memory-operand-only
-- go: Broadcast1To2
+- go: broadcast1To2
   asm: VPBROADCASTQ
   in:
   - class: vreg
@@ -405,7 +482,7 @@
     base: int
     OverwriteBase: float
 
-- go: Broadcast1To4
+- go: broadcast1To4
   asm: VPBROADCAST[BWDQ]
   in:
   - class: vreg
@@ -416,7 +493,7 @@
     lanes: 4
     base: $b
 
-- go: Broadcast1To8
+- go: broadcast1To8
   asm: VPBROADCAST[BWDQ]
   in:
   - class: vreg
@@ -427,7 +504,7 @@
     lanes: 8
     base: $b
 
-- go: Broadcast1To16
+- go: broadcast1To16
   asm: VPBROADCAST[BWDQ]
   in:
   - class: vreg
@@ -438,7 +515,7 @@
     lanes: 16
     base: $b
 
-- go: Broadcast1To32
+- go: broadcast1To32
   asm: VPBROADCAST[BWDQ]
   in:
   - class: vreg
@@ -449,7 +526,7 @@
     lanes: 32
     base: $b
 
-- go: Broadcast1To64
+- go: broadcast1To64
   asm: VPBROADCASTB
   in:
   - class: vreg
@@ -460,7 +537,7 @@
     lanes: 64
     base: $b
 
-- go: Broadcast1To4
+- go: broadcast1To4
   asm: VBROADCASTS[SD]
   in:
   - class: vreg
@@ -471,7 +548,7 @@
     lanes: 4
     base: float
 
-- go: Broadcast1To8
+- go: broadcast1To8
   asm: VBROADCASTS[SD]
   in:
   - class: vreg
@@ -482,7 +559,7 @@
     lanes: 8
     base: float
 
-- go: Broadcast1To16
+- go: broadcast1To16
   asm: VBROADCASTS[SD]
   in:
   - class: vreg
@@ -496,9 +573,6 @@
 # VPSHUFB for 128-bit byte shuffles will be picked with higher priority than VPERMB, given its lower CPU feature requirement. (It's AVX)
 - go: PermuteOrZero
   asm: VPSHUFB
-  addDoc: !string |-
-    // The lower four bits of each byte-sized index in indices select an element from x,
-    // unless the index's sign bit is set in which case zero is used instead.
   in:
   - &128any
     bits: 128
@@ -511,13 +585,6 @@
 
 - go: PermuteOrZeroGrouped
   asm: VPSHUFB
-  addDoc: !string |-
-    //
-    //   result = {x_group0[indices[0]], x_group0[indices[1]], ..., x_group1[indices[16]], x_group1[indices[17]], ...}
-    //
-    // The lower four bits of each byte-sized index in indices select an element from its corresponding group in x,
-    // unless the index's sign bit is set in which case zero is used instead.
-    // Each group is of size 128-bit.
   in:
   - &256Or512any
     bits: "256|512"
@@ -561,6 +628,26 @@
   out:
   - *256Or512any
 
+- go: permuteScalarsLo # For AVX
+  asm: VPSHUFLW
+  addDoc: !string |-
+    //
+    //   result = {x[indices[0:2]], x[indices[2:4]], x[indices[4:6]], x[indices[6:8]], x[4], x[5], x[6], x[7]}
+    //
+    // Indices is four 2-bit values packed into a byte, thus indices[0:2] is the first index.
+  in:
+  - &32x4To16x8
+    bits: 128
+    go: $t
+    elemBits: 32
+    OverwriteElementBits: 16 # XED data incorrectly specified this as 32-bit, we need to fix it.
+  - class: immediate
+    immOffset: 0
+    name: indices
+  hideMaskMethods: true
+  out:
+  - *32x4To16x8
+
 - go: permuteScalarsLo
   asm: VPSHUFLW
   addDoc: !string |-
@@ -617,6 +704,26 @@
   out:
   - *128lanes8
 
+- go: permuteScalarsHi # For AVX
+  asm: VPSHUFHW
+  addDoc: !string |-
+    //
+    //   result = {x[0], x[1], x[2], x[3], x[indices[0:2]+4], x[indices[2:4]+4], x[indices[4:6]+4], x[indices[6:8]+4]}
+    //
+    // Indices is four 2-bit values packed into a byte, thus indices[0:2] is the first index.
+  in:
+  - &32x4To16x8
+    bits: 128
+    go: $t
+    elemBits: 32
+    OverwriteElementBits: 16 # XED data incorrectly specified this as 32-bit, we need to fix it.
+  - class: immediate
+    immOffset: 0
+    name: indices
+  hideMaskMethods: true
+  out:
+  - *32x4To16x8
+
 - go: permuteScalarsHiGrouped
   asm: VPSHUFHW
   addDoc: !string |-
@@ -879,7 +986,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2F128
   operandOrder: II
   addDoc: !string |-
@@ -902,7 +1009,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2F128
   operandOrder: II
   addDoc: !string |-
@@ -926,7 +1033,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2I128
   operandOrder: II
   addDoc: !string |-
@@ -950,7 +1057,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2I128
   operandOrder: II
   addDoc: !string |-
@@ -974,7 +1081,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2I128
   operandOrder: II
   addDoc: !string |-
@@ -999,7 +1106,7 @@
   out:
   - *v
 
-- go: Select128FromPair
+- go: ConcatPermute128Scalars
   asm: VPERM2I128
   operandOrder: II
   addDoc: !string |-
@@ -1026,6 +1133,7 @@
 
 - go: ConcatShiftBytesRight
   asm: VPALIGNR
+  operandOrder: 2I
   in:
   - &uint128
     go: $t
@@ -1040,6 +1148,7 @@
 
 - go: ConcatShiftBytesRightGrouped
   asm: VPALIGNR
+  operandOrder: 2I
   in:
   - &uint256512
     go: $t