@ttsc/win32-arm64 0.33.1 → 0.34.0

package/bin/go/src/simd/archsimd/_gen/simdgen/ops/ShiftRotate/categories.yaml

@@ -1,103 +1,218 @@
 !sum
+
+- go: Shift
+  commutative: false
+  documentation: !string |-
+    // NAME shifts each element in x by the signed value of the least significant byte
+    // of y's corresponding element: positive values shift left, negative values shift right.
+- go: ShiftSaturated
+  signed: true
+  commutative: false
+  documentation: !string |-
+    // NAME shifts each element in x by the signed value of the least significant byte
+    // of y's corresponding element (positive shifts left, negative shifts right).
+    // Results are saturated to the signed range on overflow.
+- go: ShiftSaturated
+  signed: false
+  commutative: false
+  documentation: !string |-
+    // NAME shifts each element in x by the signed value of the least significant byte
+    // of y's corresponding element (positive shifts left, negative shifts right).
+    // Results are saturated to the unsigned range on overflow.
+- go: shiftLeftSaturatedConst
+  noGenericOps: "true"
+  signed: true
+  commutative: false
+  documentation: !string |-
+    // NAME performs a saturating left shift on each element in x by the constant number of bits specified by y.
+    // Results are saturated to the signed range on overflow.
+- go: shiftLeftSaturatedConst
+  noGenericOps: "true"
+  signed: false
+  commutative: false
+  documentation: !string |-
+    // NAME performs a saturating left shift on each element in x by the constant number of bits specified by y.
+    // Results are saturated to the unsigned range on overflow.
+
+- go: shiftRightNarrowConst
+  noGenericOps: "true"
+  commutative: false
+  documentation: !string |-
+    // NAME performs a right shift on each element in x by the constant number of bits
+    // and narrows the result to half the element width.
+- go: shiftLeftWidenLoConst
+  noGenericOps: "true"
+  signed: false
+  commutative: false
+  documentation: !string |-
+    // NAME performs a left shift on each unsigned low-indexed element in x by the constant number of bits
+    // and widens the result to double the element width.
+- go: shiftLeftWidenLoConst
+  noGenericOps: "true"
+  signed: true
+  commutative: false
+  documentation: !string |-
+    // NAME performs a left shift on each signed low-indexed element in x by the constant number of bits
+    // and widens the result to double the element width.
+
+# Shared ops (AMD64 + ARM64)
 - go: ShiftAllLeft
-  nameAndSizeCheck: true
-  specialLower: sftimm
   commutative: false
   documentation: !string |-
-    // NAME shifts each element to the left by y bits.
+    // NAME shifts each element of x left by y bits.
+    // If y is greater than the element width, the result is 0.
 - go: ShiftAllRight
   signed: false
-  nameAndSizeCheck: true
-  specialLower: sftimm
   commutative: false
   documentation: !string |-
-    // NAME performs an unsigned right shift on each element by y bits.
+    // NAME logically shifts each element of x right by y bits.
+    // If y is greater than the element width, the result is 0.
 - go: ShiftAllRight
   signed: true
-  specialLower: sftimm
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
-    // NAME performs a signed right shift on each element by y bits.
+    // NAME arithmetically shifts each element of x right by y bits.
+    // If y is greater than the element width, the result is 0 or -1.
 - go: shiftAllLeftConst # no APIs, only ssa ops.
   noTypes: "true"
   noGenericOps: "true"
-  SSAVariant: "const" # to avoid its name colliding with reg version of this instruction, amend this to its ssa op name.
-  nameAndSizeCheck: true
   commutative: false
 - go: shiftAllRightConst # no APIs, only ssa ops.
   noTypes: "true"
   noGenericOps: "true"
-  SSAVariant: "const"
   signed: false
-  nameAndSizeCheck: true
   commutative: false
 - go: shiftAllRightConst # no APIs, only ssa ops.
   noTypes: "true"
   noGenericOps: "true"
-  SSAVariant: "const"
   signed: true
-  nameAndSizeCheck: true
   commutative: false
 
 - go: ShiftLeft
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
-    // NAME shifts each element in x to the left by the number of bits specified in y's corresponding elements.
+    // NAME shifts x[i] left by y[i] bits.
+    // If y[i] is greater than the element width, the result is 0.
 - go: ShiftRight
   signed: false
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
-    // NAME performs an unsigned right shift on each element in x by the number of bits specified in y's corresponding elements.
+    // NAME logically shifts x[i] right by y[i] bits.
+    // If y[i] is greater than the element width, the result is 0.
 - go: ShiftRight
   signed: true
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
-    // NAME performs a signed right shift on each element in x by the number of bits specified in y's corresponding elements.
+    // NAME arithmetically shifts x[i] right by y[i] bits.
+    // If y[i] is greater than the element width, the result is 0 or -1.
 - go: RotateAllLeft
-  nameAndSizeCheck: true
+  commutative: false
+  documentation: !string |-
+    // NAME rotates each element to the left by the number of bits specified by shift.
+- go: rotateAllLeft
+  noGenericOps: "true"
   commutative: false
   documentation: !string |-
     // NAME rotates each element to the left by the number of bits specified by shift.
 - go: RotateLeft
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
     // NAME rotates each element in x to the left by the number of bits specified by y's corresponding elements.
 - go: RotateAllRight
-  nameAndSizeCheck: true
+  commutative: false
+  documentation: !string |-
+    // NAME rotates each element to the right by the number of bits specified by shift.
+- go: rotateAllRight
+  noGenericOps: "true"
   commutative: false
   documentation: !string |-
     // NAME rotates each element to the right by the number of bits specified by shift.
 - go: RotateRight
-  nameAndSizeCheck: true
   commutative: false
   documentation: !string |-
     // NAME rotates each element in x to the right by the number of bits specified by y's corresponding elements.
-- go: ShiftAllLeftConcat
-  nameAndSizeCheck: true
+# Bizzare shifts
+- go: ShiftAllLeftConcatMod16
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] left by shift%16, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift%16)
+- go: ShiftAllRightConcatMod16
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] right by shift%16, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift%16)
+- go: ShiftLeftConcatMod16
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] left by shift[i]%16, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift[i]%16)
+- go: ShiftRightConcatMod16
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] right by shift[i]%16, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift[i]%16)
+- go: ShiftAllLeftConcatMod32
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] left by shift%32, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift%32)
+- go: ShiftAllRightConcatMod32
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] right by shift%32, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift%32)
+- go: ShiftLeftConcatMod32
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] left by shift[i]%32, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift[i]%32)
+- go: ShiftRightConcatMod32
+  commutative: false
+  documentation: !string |-
+    // NAME shifts x[i] right by shift[i]%32, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift[i]%32)
+- go: ShiftAllLeftConcatMod64
   commutative: false
   documentation: !string |-
-    // NAME shifts each element of x to the left by the number of bits specified by
-    // shift (only the lower 5 bits are used), and then copies the upper bits of y to the emptied lower bits of the shifted x.
-- go: ShiftAllRightConcat
-  nameAndSizeCheck: true
+    // NAME shifts x[i] left by shift%64, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift%64)
+- go: ShiftAllRightConcatMod64
   commutative: false
   documentation: !string |-
-    // NAME shifts each element of x to the right by the number of bits specified by
-    // shift (only the lower 5 bits are used), and then copies the lower bits of y to the emptied upper bits of the shifted x.
-- go: ShiftLeftConcat
-  nameAndSizeCheck: true
+    // NAME shifts x[i] right by shift%64, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift%64)
+- go: ShiftLeftConcatMod64
   commutative: false
   documentation: !string |-
-    // NAME shifts each element of x to the left by the number of bits specified by the
-    // corresponding elements in y (only the lower 5 bits are used), and then copies the upper bits of z to the emptied lower bits of the shifted x.
-- go: ShiftRightConcat
-  nameAndSizeCheck: true
+    // NAME shifts x[i] left by shift[i]%64, filing any empted lower bits
+    // with the high bits of y[i].
+    //
+    //    z[i] = concat(x[i], y[i]) << (shift[i]%64)
+- go: ShiftRightConcatMod64
   commutative: false
   documentation: !string |-
-    // NAME shifts each element of x to the right by the number of bits specified by the
-    // corresponding elements in y (only the lower 5 bits are used), and then copies the lower bits of z to the emptied upper bits of the shifted x.
+    // NAME shifts x[i] right by shift[i]%64, filling any emptied upper bits
+    // with the low bits of y[i].
+    //
+    //    z[i] = concat(y[i], x[i]) >> (shift[i]%64)

package/bin/go/src/simd/archsimd/_gen/simdgen/ops/ShiftRotate/go_amd64.yaml

@@ -0,0 +1,299 @@
+!sum
+# Integers
+# ShiftAll*
+- go: ShiftAllLeft
+  specialLower: sftimm
+  asm: "VPSLL[WDQ]"
+  in:
+  - &any
+    go: $t
+  - &vecAsScalar64
+    go: "Uint.*"
+    treatLikeAScalarOfSize: 64
+    name: shift
+  out:
+  - *any
+- go: ShiftAllRight
+  signed: false
+  specialLower: sftimm
+  asm: "VPSRL[WDQ]"
+  in:
+  - &uint
+    go: $t
+    base: uint
+  - *vecAsScalar64
+  out:
+  - *uint
+- go: ShiftAllRight
+  signed: true
+  specialLower: sftimm
+  asm: "VPSRA[WDQ]"
+  in:
+  - &int
+    go: $t
+    base: int
+  - *vecAsScalar64
+  out:
+  - *int
+
+- go: shiftAllLeftConst
+  SSAVariant: "const" # to avoid its name colliding with reg version of this instruction, amend this to its ssa op name.
+  nameAndSizeCheck: true
+  specialLower: "if a==0 => x"
+  asm: "VPSLL[WDQ]"
+  in:
+  - *any
+  - &imm
+    class: immediate
+    immOffset: 0
+  out:
+  - *any
+- go: shiftAllRightConst
+  SSAVariant: "const" # to avoid its name colliding with reg version of this instruction, amend this to its ssa op name.
+  nameAndSizeCheck: true
+  specialLower: "if a==0 => x"
+  asm: "VPSRA[WDQ]"
+  in:
+  - *int
+  - *imm
+  out:
+  - *int
+- go: shiftAllRightConst
+  SSAVariant: "const" # to avoid its name colliding with reg version of this instruction, amend this to its ssa op name.
+  nameAndSizeCheck: true
+  specialLower: "if a==0 => x"
+  asm: "VPSRL[WDQ]"
+  in:
+  - *uint
+  - *imm
+  out:
+  - *uint
+
+# Shift* (variable)
+- go: ShiftLeft
+  nameAndSizeCheck: true
+  asm: "VPSLLV[WD]"
+  in:
+  - go: $t
+  - go: $t
+    overwriteBase: uint
+    name: shift
+  out:
+  - go: $t
+# XED data of VPSLLVQ marks the element bits 32 which is off to the actual semantic, we need to overwrite
+# it to 64.
+- go: ShiftLeft
+  nameAndSizeCheck: true
+  asm: "VPSLLVQ"
+  in:
+  - go: $t
+    overwriteElementBits: 64
+  - go: $t
+    overwriteElementBits: 64
+    overwriteBase: uint
+    name: shift
+  out:
+  - go: $t
+    overwriteElementBits: 64
+- go: ShiftRight
+  signed: false
+  nameAndSizeCheck: true
+  asm: "VPSRLV[WD]"
+  in:
+  - go: $t
+    base: uint
+  - go: $t
+    overwriteBase: uint
+    name: shift
+  out:
+  - go: $t
+    base: uint
+# XED data of VPSRLVQ needs the same overwrite as VPSLLVQ.
+- go: ShiftRight
+  signed: false
+  nameAndSizeCheck: true
+  asm: "VPSRLVQ"
+  in:
+  - go: $t
+    base: uint
+    overwriteElementBits: 64
+  - go: $t
+    overwriteElementBits: 64
+    overwriteBase: uint
+    name: shift
+  out:
+  - go: $t
+    base: uint
+    overwriteElementBits: 64
+- go: ShiftRight
+  signed: true
+  nameAndSizeCheck: true
+  asm: "VPSRAV[WDQ]"
+  in:
+  - go: $t
+    base: int
+  - go: $t
+    overwriteBase: uint
+    name: shift
+  out:
+  - go: $t
+    base: int
+
+# Rotate
+- go: rotateAllLeft
+  nameAndSizeCheck: true
+  asm: "VPROL[DQ]"
+  in:
+  - *any
+  - &pureImm
+    class: immediate
+    immOffset: 0
+    name: shift
+  out:
+  - *any
+- go: rotateAllRight
+  nameAndSizeCheck: true
+  asm: "VPROR[DQ]"
+  in:
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: RotateLeft
+  nameAndSizeCheck: true
+  asm: "VPROLV[DQ]"
+  in:
+  - *any
+  - *any
+  out:
+  - *any
+- go: RotateRight
+  nameAndSizeCheck: true
+  asm: "VPRORV[DQ]"
+  in:
+  - *any
+  - *any
+  out:
+  - *any
+
+# Bizzare shifts.
+- go: ShiftAllLeftConcatMod16
+  nameAndSizeCheck: true
+  asm: "VPSHLDW"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftAllRightConcatMod16
+  nameAndSizeCheck: true
+  asm: "VPSHRDW"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftLeftConcatMod16
+  nameAndSizeCheck: true
+  asm: "VPSHLDVW"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t
+- go: ShiftRightConcatMod16
+  nameAndSizeCheck: true
+  asm: "VPSHRDVW"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t
+- go: ShiftAllLeftConcatMod32
+  nameAndSizeCheck: true
+  asm: "VPSHLDD"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftAllRightConcatMod32
+  nameAndSizeCheck: true
+  asm: "VPSHRDD"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftLeftConcatMod32
+  nameAndSizeCheck: true
+  asm: "VPSHLDVD"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t
+- go: ShiftRightConcatMod32
+  nameAndSizeCheck: true
+  asm: "VPSHRDVD"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t
+- go: ShiftAllLeftConcatMod64
+  nameAndSizeCheck: true
+  asm: "VPSHLDQ"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftAllRightConcatMod64
+  nameAndSizeCheck: true
+  asm: "VPSHRDQ"
+  operandOrder: 2I
+  in:
+  - *any
+  - *any
+  - *pureImm
+  out:
+  - *any
+- go: ShiftLeftConcatMod64
+  nameAndSizeCheck: true
+  asm: "VPSHLDVQ"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t
+- go: ShiftRightConcatMod64
+  nameAndSizeCheck: true
+  asm: "VPSHRDVQ"
+  in:
+  - go: $t
+  - go: $t
+  - base: uint
+    name: shift
+  out:
+  - go: $t