[dev.simd] simd, cmd/compile: add more element types for Select128FromPair

Also includes a comment cleanup pass.
Fixed NAME processing for additional documentation.

Change-Id: Ide5b60c17ddbf3c6eafd20147981c59493fc8133
Reviewed-on: https://go-review.googlesource.com/c/go/+/722180
Reviewed-by: Junyang Shao <shaojunyang@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

Author: dr2chase

src/cmd/compile/internal/ssa/_gen/simdAMD64.rules

@@ -941,8 +941,12 @@
 (ScaleFloat64x8 ...) => (VSCALEFPD512 ...)
 (Select128FromPairFloat32x8 ...) => (VPERM2F128256 ...)
 (Select128FromPairFloat64x4 ...) => (VPERM2F128256 ...)
+(Select128FromPairInt8x32 ...) => (VPERM2I128256 ...)
+(Select128FromPairInt16x16 ...) => (VPERM2I128256 ...)
 (Select128FromPairInt32x8 ...) => (VPERM2I128256 ...)
 (Select128FromPairInt64x4 ...) => (VPERM2I128256 ...)
+(Select128FromPairUint8x32 ...) => (VPERM2I128256 ...)
+(Select128FromPairUint16x16 ...) => (VPERM2I128256 ...)
 (Select128FromPairUint32x8 ...) => (VPERM2I128256 ...)
 (Select128FromPairUint64x4 ...) => (VPERM2I128256 ...)
 (SetElemFloat32x4 ...) => (VPINSRD128 ...)

src/cmd/compile/internal/ssa/_gen/simdgenericOps.go

@@ -351,7 +351,7 @@ func ({{.Op1NameAndType "x"}}) {{.Go}}({{.Op2NameAndType "y"}}, {{.ImmName}} uin
 {{if .Documentation}}{{.Documentation}}
 //{{end}}
 // {{.ImmName}} result in better performance when they are constants, non-constant values will be translated into a jump table.
-// {{.ImmName}} should be between 0 and 3, inclusive; other values will result in a runtime panic.
+// {{.ImmName}} should be between 0 and 3, inclusive; other values may result in a runtime panic.
 //
 // Asm: {{.Asm}}, CPU Feature: {{.CPUFeature}}
 func ({{.Op1NameAndType "x"}}) {{.Go}}({{.ImmName}} uint8, {{.Op2NameAndType "y"}}) {{.GoType}}

src/cmd/compile/internal/ssa/opGen.go

@@ -98,6 +98,8 @@ func (o *Operation) SkipMaskedMethod() bool {
 	return false
 }
 
+var reForName = regexp.MustCompile(`\bNAME\b`)
+
 func (o *Operation) DecodeUnified(v *unify.Value) error {
 	if err := v.Decode(&o.rawOperation); err != nil {
 		return err
@@ -117,7 +119,7 @@ func (o *Operation) DecodeUnified(v *unify.Value) error {
 	} else {
 		o.Documentation = "// UNDOCUMENTED"
 	}
-	o.Documentation = regexp.MustCompile(`\bNAME\b`).ReplaceAllString(o.Documentation, o.Go)
+	o.Documentation = reForName.ReplaceAllString(o.Documentation, o.Go)
 	if isMasked {
 		o.Documentation += "\n//\n// This operation is applied selectively under a write mask."
 		// Suppress generic op and method declaration for exported methods, if a mask is present.
@@ -128,7 +130,7 @@ func (o *Operation) DecodeUnified(v *unify.Value) error {
 		}
 	}
 	if o.rawOperation.AddDoc != nil {
-		o.Documentation += "\n" + *o.rawOperation.AddDoc
+		o.Documentation += "\n" + reForName.ReplaceAllString(*o.rawOperation.AddDoc, o.Go)
 	}
 
 	o.In = append(o.rawOperation.In, o.rawOperation.InVariant...)

src/cmd/compile/internal/ssa/rewriteAMD64.go

@@ -135,7 +135,7 @@
     // NAME concatenates selected elements from x and y into the lower and upper
     // halves of the output.  The selection is chosen by the constant parameter h1h0l1l0
     // where each {h,l}{1,0} is two bits specify which element from y or x to select.
-    // For example, {0,1,2,3}.concatSelectedConstant(0b_11_01_00_10, {4,5,6,7}) returns
+    // For example, {0,1,2,3}.NAME(0b_11_01_00_10, {4,5,6,7}) returns
     // {2, 0, 5, 7} (don't forget that the binary constant is written big-endian).
 
 - go: concatSelectedConstant
@@ -196,9 +196,12 @@
     // The selection is chosen by the constant parameter h1h0l1l0
     // where each {h,l}{1,0} is two bits specifying which element from y or x to select.
     // For example,
-    // {0,1,2,3,8,9,10,11, 20,21,22,23,28,29,210,211}.NAME(
-    //             0b_11_01_00_10, {4,5,6,7,12,13,14,15, 24,25,26,27,212,213,214,215})
+    //
+    //   {0,1,2,3,8,9,10,11, 20,21,22,23,28,29,210,211}.NAME(
+    //    0b_11_01_00_10, {4,5,6,7,12,13,14,15, 24,25,26,27,212,213,214,215})
+    //
     // returns {2,0,5,7,10,8,13,15, 22,20,25,27,210,28,213,215}
+    //
     // (don't forget that the binary constant is written big-endian).
 
 - go: concatSelectedConstantGrouped
@@ -214,7 +217,7 @@
     // subvectors of x and y.
     //
     // For example {4,5,8,9,12,13,16,17}.NAME(0b11_00_11_10, {6,7,10,11,14,15,18,19})
-    // returns {4,7,9,11,12,14,17,19}; bit 0 is zero, selecting element 0 from x's 
+    // returns {4,7,9,11,12,14,17,19}; bit 0 is zero, selecting element 0 from x's
     // least 128-bits (4), then 1, selects the element 1 from y's least 128-bits (7),
     // then 1, selecting element 1 from x's next 128 bits (9), then 1,
     // selecting element 1 from y's upper 128 bits (11).  The next two 0 bits select
@@ -227,9 +230,8 @@
   commutative: false
   documentation: !string |-
     // NAME treats the 256-bit vectors x and y as a single vector of four
-    // 128-bit elements, and returns a 256-bit result formed by 
+    // 128-bit elements, and returns a 256-bit result formed by
     // concatenating the two elements specified by lo and hi.
-    // For example, {4,5}.NAME(3,0,{6,7}) returns {7,4}.
 
 - go: ConcatShiftBytesRight
   commutative: false

src/cmd/compile/internal/ssagen/simdintrinsics.go

@@ -837,6 +837,12 @@
 - go: Select128FromPair
   asm: VPERM2F128
   operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {40, 41, 50, 51}.NAME(3, 0, {60, 61, 70, 71})
+    //
+    // returns {70, 71, 40, 41}.
   in:
   - &v
     go: $t
@@ -854,6 +860,12 @@
 - go: Select128FromPair
   asm: VPERM2F128
   operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {40, 41, 42, 43, 50, 51, 52, 53}.NAME(3, 0, {60, 61, 62, 63, 70, 71, 72, 73})
+    //
+    // returns {70, 71, 72, 73, 40, 41, 42, 43}.
   in:
   - &v
     go: $t
@@ -872,6 +884,12 @@
 - go: Select128FromPair
   asm: VPERM2I128
   operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {40, 41, 50, 51}.NAME(3, 0, {60, 61, 70, 71})
+    //
+    // returns {70, 71, 40, 41}.
   in:
   - &v
     go: $t
@@ -890,6 +908,12 @@
 - go: Select128FromPair
   asm: VPERM2I128
   operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {40, 41, 42, 43, 50, 51, 52, 53}.NAME(3, 0, {60, 61, 62, 63, 70, 71, 72, 73})
+    //
+    // returns {70, 71, 72, 73, 40, 41, 42, 43}.
   in:
   - &v
     go: $t
@@ -905,6 +929,56 @@
   out:
   - *v
 
+- go: Select128FromPair
+  asm: VPERM2I128
+  operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {40, 41, 42, 43, 44, 45, 46, 47, 50, 51, 52, 53, 54, 55, 56, 57}.NAME(3, 0,
+    //    {60, 61, 62, 63, 64, 65, 66, 67, 70, 71, 72, 73, 74, 75, 76, 77})
+    //
+    // returns {70, 71, 72, 73, 74, 75, 76, 77, 40, 41, 42, 43, 44, 45, 46, 47}.
+  in:
+  - &v
+    go: $t
+    class: vreg
+    base: int|uint
+    bits: 256
+    OverwriteElementBits: 16
+  - *v
+  - class: immediate
+    immOffset: 0
+    name: "lo, hi"
+  inVariant: []
+  out:
+  - *v
+
+- go: Select128FromPair
+  asm: VPERM2I128
+  operandOrder: II
+  addDoc: !string |-
+    // For example,
+    //
+    //   {0x40, 0x41, ..., 0x4f, 0x50, 0x51, ..., 0x5f}.NAME(3, 0,
+    //        {0x60, 0x61, ..., 0x6f, 0x70, 0x71, ..., 0x7f})
+    //
+    // returns {0x70, 0x71, ..., 0x7f, 0x40, 0x41, ..., 0x4f}.
+  in:
+  - &v
+    go: $t
+    class: vreg
+    base: int|uint
+    bits: 256
+    OverwriteElementBits: 8
+  - *v
+  - class: immediate
+    immOffset: 0
+    name: "lo, hi"
+  inVariant: []
+  out:
+  - *v
+
 - go: ConcatShiftBytesRight
   asm: VPALIGNR
   in:
@@ -930,4 +1004,3 @@
     immOffset: 0
   out:
   - *uint256512
-