runtime/internal/atomic: Use power5 compatible instructions for ppc64

This modifies a recent performance improvement to the
And8 and Or8 atomic functions which required both ppc64le
and ppc64 to use power8 instructions. Since then it was
decided that ppc64 (BE) should work for power5 and later.
This change uses instructions compatible with power5 for
ppc64 and uses power8 for ppc64le.

Fixes #16004

Change-Id: I623c75e8e6fd1fa063a53d250d86cdc9d0890dc7
Reviewed-on: https://go-review.googlesource.com/24181Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Andrew Gerrand <adg@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/runtime/internal/atomic/asm_ppc64x.s

@@ -161,28 +161,71 @@ TEXT runtime∕internal∕atomic·Store64(SB), NOSPLIT, $0-16
 	MOVD	R4, 0(R3)
 	RET
 
-// void	runtime∕internal∕atomic·Or8(byte volatile*, byte);
+// void runtime∕internal∕atomic·Or8(byte volatile*, byte);
 TEXT runtime∕internal∕atomic·Or8(SB), NOSPLIT, $0-9
 	MOVD	ptr+0(FP), R3
 	MOVBZ	val+8(FP), R4
+#ifdef  GOARCH_ppc64
+	// Align ptr down to 4 bytes so we can use 32-bit load/store.
+	// R5 = (R3 << 0) & ~3
+	RLDCR	$0, R3, $~3, R5
+	// Compute val shift.
+	// Big endian.  ptr = ptr ^ 3
+	XOR	$3, R3
+	// R6 = ((ptr & 3) * 8) = (ptr << 3) & (3*8)
+	RLDC	$3, R3, $(3*8), R6
+	// Shift val for aligned ptr.  R4 = val << R6
+	SLD	R6, R4, R4
+	SYNC
+
+again:
+	LWAR	(R5), R6
+	OR	R4, R6
+	STWCCC	R6, (R5)
+	BNE	again
+#else
 	SYNC
 again:
 	LBAR	(R3), R6
 	OR	R4, R6
 	STBCCC	R6, (R3)
 	BNE	again
+#endif
 	ISYNC
 	RET
 
-// void	runtime∕internal∕atomic·And8(byte volatile*, byte);
+// void runtime∕internal∕atomic·And8(byte volatile*, byte);
 TEXT runtime∕internal∕atomic·And8(SB), NOSPLIT, $0-9
 	MOVD	ptr+0(FP), R3
 	MOVBZ	val+8(FP), R4
+#ifdef  GOARCH_ppc64
+	// Align ptr down to 4 bytes so we can use 32-bit load/store.
+	// R5 = (R3 << 0) & ~3
+	RLDCR	$0, R3, $~3, R5
+	// Compute val shift.
+	// Big endian.  ptr = ptr ^ 3
+	XOR	$3, R3
+	// R6 = ((ptr & 3) * 8) = (ptr << 3) & (3*8)
+	RLDC	$3, R3, $(3*8), R6
+	// Shift val for aligned ptr.  R4 = val << R6 | ^(0xFF << R6)
+	MOVD	$0xFF, R7
+	SLD	R6, R4
+	SLD	R6, R7
+	XOR	$-1, R7
+	OR	R7, R4
 	SYNC
 again:
-	LBAR	(R3), R6
+	LWAR	(R5), R6
 	AND	R4, R6
-	STBCCC	R6, (R3)
+	STWCCC	R6, (R5)
+	BNE	again
+#else
+	SYNC
+again:
+	LBAR	(R3),R6
+	AND	R4,R6
+	STBCCC	R6,(R3)
 	BNE	again
+#endif
 	ISYNC
 	RET