cmd/compile: implement non-constant rotates

Makes math/bits.Rotate{Left,Right} fast on amd64.

name              old time/op  new time/op  delta
RotateLeft-12     7.42ns ± 6%  5.45ns ± 6%  -26.54%   (p=0.000 n=9+10)
RotateLeft8-12    4.77ns ± 5%  3.42ns ± 7%  -28.25%   (p=0.000 n=8+10)
RotateLeft16-12   4.82ns ± 8%  3.40ns ± 7%  -29.36%  (p=0.000 n=10+10)
RotateLeft32-12   4.87ns ± 7%  3.48ns ± 7%  -28.51%    (p=0.000 n=8+9)
RotateLeft64-12   5.23ns ±10%  3.35ns ± 6%  -35.97%   (p=0.000 n=9+10)
RotateRight-12    7.59ns ± 8%  5.71ns ± 1%  -24.72%   (p=0.000 n=10+8)
RotateRight8-12   4.98ns ± 7%  3.36ns ± 9%  -32.55%  (p=0.000 n=10+10)
RotateRight16-12  5.12ns ± 2%  3.45ns ± 5%  -32.62%  (p=0.000 n=10+10)
RotateRight32-12  4.80ns ± 6%  3.42ns ±16%  -28.68%  (p=0.000 n=10+10)
RotateRight64-12  4.78ns ± 6%  3.42ns ± 6%  -28.50%  (p=0.000 n=10+10)

Update #18940

Change-Id: Ie79fb5581c489ed4d3b859314c5e669a134c119b
Reviewed-on: https://go-review.googlesource.com/39711
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

src/cmd/compile/internal/gc/asm_test.go

@@ -800,6 +800,63 @@ var linuxAMD64Tests = []*asmTest{
 		}`,
 		[]string{"\tCMPQ\t[A-Z]"},
 	},
+	// Non-constant rotate
+	{
+		`func rot64l(x uint64, y int) uint64 {
+			z := uint(y & 63)
+			return x << z | x >> (64-z)
+		}`,
+		[]string{"\tROLQ\t"},
+	},
+	{
+		`func rot64r(x uint64, y int) uint64 {
+			z := uint(y & 63)
+			return x >> z | x << (64-z)
+		}`,
+		[]string{"\tRORQ\t"},
+	},
+	{
+		`func rot32l(x uint32, y int) uint32 {
+			z := uint(y & 31)
+			return x << z | x >> (32-z)
+		}`,
+		[]string{"\tROLL\t"},
+	},
+	{
+		`func rot32r(x uint32, y int) uint32 {
+			z := uint(y & 31)
+			return x >> z | x << (32-z)
+		}`,
+		[]string{"\tRORL\t"},
+	},
+	{
+		`func rot16l(x uint16, y int) uint16 {
+			z := uint(y & 15)
+			return x << z | x >> (16-z)
+		}`,
+		[]string{"\tROLW\t"},
+	},
+	{
+		`func rot16r(x uint16, y int) uint16 {
+			z := uint(y & 15)
+			return x >> z | x << (16-z)
+		}`,
+		[]string{"\tRORW\t"},
+	},
+	{
+		`func rot8l(x uint8, y int) uint8 {
+			z := uint(y & 7)
+			return x << z | x >> (8-z)
+		}`,
+		[]string{"\tROLB\t"},
+	},
+	{
+		`func rot8r(x uint8, y int) uint8 {
+			z := uint(y & 7)
+			return x >> z | x << (8-z)
+		}`,
+		[]string{"\tRORB\t"},
+	},
 }
 
 var linux386Tests = []*asmTest{