simd/archsimd: add Grouped for 256- and 512-bit SaturateTo(U)Int16Concat, and fix type

They operate on 128-bit groups, so name them Grouped to be clear,
and consistent with other grouped operations. Reword the
documentation, mention the grouping only for grouped versions.

Also, SaturateToUnt16Concat(Grouped) is a signed int32 to unsigned
uint16 saturated conversion. The receiver and the parameter should
be signed. The result remains unsigned.

Change-Id: I30e28bc05e07f5c28214c9c6d9d201cbbb183468
Reviewed-on: https://go-review.googlesource.com/c/go/+/731501
Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>

4 files changed, 59 insertions, 22 deletions

diff --git a/src/simd/archsimd/_gen/simdgen/godefs.go b/src/simd/archsimd/_gen/simdgen/godefs.go
index 3ac74264e8..71cae158f7 100644
--- a/src/simd/archsimd/_gen/simdgen/godefs.go
+++ b/src/simd/archsimd/_gen/simdgen/godefs.go
@@ -142,7 +142,7 @@ func (o *Operation) DecodeUnified(v *unify.Value) error {
 		outLanes := o.Out[0].Lanes
 		if inLanes != nil && outLanes != nil && *inLanes < *outLanes {
 			if (strings.Contains(o.Go, "Saturate") || strings.Contains(o.Go, "Truncate")) &&
-				!strings.HasSuffix(o.Go, "Concat") {
+				!strings.Contains(o.Go, "Concat") {
 				o.Documentation += "\n// Results are packed to low elements in the returned vector, its upper elements are zeroed."
 			}
 		}
diff --git a/src/simd/archsimd/_gen/simdgen/ops/Converts/categories.yaml b/src/simd/archsimd/_gen/simdgen/ops/Converts/categories.yaml
index 1e2a6a9b69..88e4840920 100644
--- a/src/simd/archsimd/_gen/simdgen/ops/Converts/categories.yaml
+++ b/src/simd/archsimd/_gen/simdgen/ops/Converts/categories.yaml
@@ -64,7 +64,7 @@
   regexpTag: "convert"
   documentation: !string |-
     // NAME truncates element values to int16.
-- go: "SaturateToInt16(Concat)?"
+- go: "SaturateToInt16(Concat(Grouped)?)?"
   commutative: false
   regexpTag: "convert"
   documentation: !string |-
@@ -109,7 +109,7 @@
   regexpTag: "convert"
   documentation: !string |-
     // NAME truncates element values to uint16.
-- go: "SaturateToUint16(Concat)?"
+- go: "SaturateToUint16(Concat(Grouped)?)?"
   commutative: false
   regexpTag: "convert"
   documentation: !string |-
diff --git a/src/simd/archsimd/_gen/simdgen/ops/Converts/go.yaml b/src/simd/archsimd/_gen/simdgen/ops/Converts/go.yaml
index 1d688b434d..f436be23ef 100644
--- a/src/simd/archsimd/_gen/simdgen/ops/Converts/go.yaml
+++ b/src/simd/archsimd/_gen/simdgen/ops/Converts/go.yaml
@@ -446,22 +446,48 @@
   asm: "VPACKSSDW"
   addDoc: &satDocConcat
     !string |-
+    // The converted elements from x will be packed to the lower part of the result vector,
+    // the converted elements from y will be packed to the upper part of the result vector.
+  in:
+    - base: int
+    - base: int
+  out:
+    - base: int
+      bits: 128
+- go: SaturateToInt16ConcatGrouped
+  regexpTag: "convert"
+  asm: "VPACKSSDW"
+  addDoc: &satDocConcatGrouped
+    !string |-
     // With each 128-bit as a group:
-    // The converted group from the first input vector will be packed to the lower part of the result vector,
-    // the converted group from the second input vector will be packed to the upper part of the result vector.
+    // The converted elements from x will be packed to the lower part of the group in the result vector,
+    // the converted elements from y will be packed to the upper part of the group in the result vector.
   in:
     - base: int
     - base: int
   out:
     - base: int
+      bits: 256|512
 - go: SaturateToUint16Concat
   regexpTag: "convert"
   asm: "VPACKUSDW"
+  addDoc: *satDocConcat
   in:
+    - base: int
+    - base: int
+  out:
     - base: uint
-    - base: uint
+      bits: 128
+- go: SaturateToUint16ConcatGrouped
+  regexpTag: "convert"
+  asm: "VPACKUSDW"
+  addDoc: *satDocConcatGrouped
+  in:
+    - base: int
+    - base: int
   out:
     - base: uint
+      bits: 256|512
 
 # low-part only conversions.
 # uint8->uint16
diff --git a/src/simd/archsimd/ops_amd64.go b/src/simd/archsimd/ops_amd64.go
index 304c0c0796..2a8a6bd4c6 100644
--- a/src/simd/archsimd/ops_amd64.go
+++ b/src/simd/archsimd/ops_amd64.go
@@ -5418,28 +5418,29 @@ func (x Int64x8) SaturateToInt16() Int16x8
 /* SaturateToInt16Concat */
 
 // SaturateToInt16Concat converts element values to int16 with signed saturation.
-// With each 128-bit as a group:
-// The converted group from the first input vector will be packed to the lower part of the result vector,
-// the converted group from the second input vector will be packed to the upper part of the result vector.
+// The converted elements from x will be packed to the lower part of the result vector,
+// the converted elements from y will be packed to the upper part of the result vector.
 //
 // Asm: VPACKSSDW, CPU Feature: AVX
 func (x Int32x4) SaturateToInt16Concat(y Int32x4) Int16x8
 
-// SaturateToInt16Concat converts element values to int16 with signed saturation.
+/* SaturateToInt16ConcatGrouped */
+
+// SaturateToInt16ConcatGrouped converts element values to int16 with signed saturation.
 // With each 128-bit as a group:
-// The converted group from the first input vector will be packed to the lower part of the result vector,
-// the converted group from the second input vector will be packed to the upper part of the result vector.
+// The converted elements from x will be packed to the lower part of the group in the result vector,
+// the converted elements from y will be packed to the upper part of the group in the result vector.
 //
 // Asm: VPACKSSDW, CPU Feature: AVX2
-func (x Int32x8) SaturateToInt16Concat(y Int32x8) Int16x16
+func (x Int32x8) SaturateToInt16ConcatGrouped(y Int32x8) Int16x16
 
-// SaturateToInt16Concat converts element values to int16 with signed saturation.
+// SaturateToInt16ConcatGrouped converts element values to int16 with signed saturation.
 // With each 128-bit as a group:
-// The converted group from the first input vector will be packed to the lower part of the result vector,
-// the converted group from the second input vector will be packed to the upper part of the result vector.
+// The converted elements from x will be packed to the lower part of the group in the result vector,
+// the converted elements from y will be packed to the upper part of the group in the result vector.
 //
 // Asm: VPACKSSDW, CPU Feature: AVX512
-func (x Int32x16) SaturateToInt16Concat(y Int32x16) Int16x32
+func (x Int32x16) SaturateToInt16ConcatGrouped(y Int32x16) Int16x32
 
 /* SaturateToInt32 */
 
@@ -5550,19 +5551,29 @@ func (x Uint64x8) SaturateToUint16() Uint16x8
 /* SaturateToUint16Concat */
 
 // SaturateToUint16Concat converts element values to uint16 with unsigned saturation.
+// The converted elements from x will be packed to the lower part of the result vector,
+// the converted elements from y will be packed to the upper part of the result vector.
 //
 // Asm: VPACKUSDW, CPU Feature: AVX
-func (x Uint32x4) SaturateToUint16Concat(y Uint32x4) Uint16x8
+func (x Int32x4) SaturateToUint16Concat(y Int32x4) Uint16x8
 
-// SaturateToUint16Concat converts element values to uint16 with unsigned saturation.
+/* SaturateToUint16ConcatGrouped */
+
+// SaturateToUint16ConcatGrouped converts element values to uint16 with unsigned saturation.
+// With each 128-bit as a group:
+// The converted elements from x will be packed to the lower part of the group in the result vector,
+// the converted elements from y will be packed to the upper part of the group in the result vector.
 //
 // Asm: VPACKUSDW, CPU Feature: AVX2
-func (x Uint32x8) SaturateToUint16Concat(y Uint32x8) Uint16x16
+func (x Int32x8) SaturateToUint16ConcatGrouped(y Int32x8) Uint16x16
 
-// SaturateToUint16Concat converts element values to uint16 with unsigned saturation.
+// SaturateToUint16ConcatGrouped converts element values to uint16 with unsigned saturation.
+// With each 128-bit as a group:
+// The converted elements from x will be packed to the lower part of the group in the result vector,
+// the converted elements from y will be packed to the upper part of the group in the result vector.
 //
 // Asm: VPACKUSDW, CPU Feature: AVX512
-func (x Uint32x16) SaturateToUint16Concat(y Uint32x16) Uint16x32
+func (x Int32x16) SaturateToUint16ConcatGrouped(y Int32x16) Uint16x32
 
 /* SaturateToUint32 */