[dev.garbage] runtime: eliminate heapBitsSetTypeNoScan

It's no longer necessary to maintain the bitmap of noscan objects
since we now use the span metadata to determine that they're noscan
instead of the bitmap.

The combined effect of segregating noscan spans and the follow-on
optimizations is almost no effect on the go1 benchmarks and a 1.19%
improvement in the garbage benchmark:

name              old time/op  new time/op  delta
XBenchGarbage-12  2.13ms ± 2%  2.11ms ± 1%  -1.19%  (p=0.000 n=19+20)

name                      old time/op    new time/op    delta
BinaryTree17-12              2.42s ± 1%     2.41s ± 1%  -0.50%  (p=0.001 n=20+17)
Fannkuch11-12                2.14s ± 0%     2.12s ± 0%  -0.91%  (p=0.000 n=20+17)
FmtFprintfEmpty-12          45.2ns ± 0%    45.2ns ± 1%    ~     (p=0.677 n=16+19)
FmtFprintfString-12          131ns ± 0%     132ns ± 1%  +0.57%  (p=0.000 n=16+20)
FmtFprintfInt-12             126ns ± 1%     126ns ± 0%    ~     (p=0.078 n=18+16)
FmtFprintfIntInt-12          199ns ± 0%     195ns ± 0%  -2.19%  (p=0.000 n=14+20)
FmtFprintfPrefixedInt-12     196ns ± 1%     196ns ± 0%    ~     (p=0.155 n=19+16)
FmtFprintfFloat-12           254ns ± 0%     253ns ± 0%  -0.50%  (p=0.000 n=14+19)
FmtManyArgs-12               803ns ± 1%     798ns ± 0%  -0.71%  (p=0.000 n=18+19)
GobDecode-12                7.11ms ± 1%    7.07ms ± 1%  -0.50%  (p=0.024 n=18+19)
GobEncode-12                5.87ms ± 0%    5.86ms ± 1%    ~     (p=0.113 n=19+20)
Gzip-12                      218ms ± 1%     218ms ± 1%    ~     (p=0.879 n=19+20)
Gunzip-12                   37.2ms ± 0%    37.3ms ± 0%  +0.14%  (p=0.047 n=19+20)
HTTPClientServer-12         80.5µs ± 6%    82.8µs ± 8%  +2.91%  (p=0.008 n=19+20)
JSONEncode-12               15.4ms ± 1%    15.4ms ± 1%  -0.32%  (p=0.003 n=18+19)
JSONDecode-12               55.1ms ± 1%    53.0ms ± 1%  -3.87%  (p=0.000 n=18+20)
Mandelbrot200-12            4.08ms ± 1%    4.10ms ± 1%  +0.34%  (p=0.001 n=19+17)
GoParse-12                  3.20ms ± 1%    3.21ms ± 1%    ~     (p=0.138 n=19+19)
RegexpMatchEasy0_32-12      70.6ns ± 2%    70.4ns ± 2%    ~     (p=0.343 n=20+20)
RegexpMatchEasy0_1K-12       240ns ± 0%     242ns ± 2%  +0.88%  (p=0.000 n=15+20)
RegexpMatchEasy1_32-12      70.5ns ± 1%    70.2ns ± 3%    ~     (p=0.053 n=18+20)
RegexpMatchEasy1_1K-12       374ns ± 1%     374ns ± 1%    ~     (p=0.705 n=20+19)
RegexpMatchMedium_32-12      108ns ± 1%     108ns ± 1%    ~     (p=0.854 n=20+19)
RegexpMatchMedium_1K-12     33.5µs ± 1%    33.6µs ± 2%    ~     (p=0.897 n=18+20)
RegexpMatchHard_32-12       1.76µs ± 1%    1.75µs ± 1%    ~     (p=0.771 n=18+18)
RegexpMatchHard_1K-12       52.8µs ± 1%    52.8µs ± 1%    ~     (p=0.678 n=17+19)
Revcomp-12                   381ms ± 1%     380ms ± 0%    ~     (p=0.320 n=20+16)
Template-12                 65.6ms ± 1%    65.1ms ± 2%  -0.75%  (p=0.003 n=20+20)
TimeParse-12                 324ns ± 1%     326ns ± 1%  +0.72%  (p=0.000 n=18+18)
TimeFormat-12                342ns ± 0%     343ns ± 1%  +0.22%  (p=0.004 n=15+18)
[Geo mean]                  52.4µs         52.3µs       -0.18%

Change-Id: Ic77faaa15cdac3bfbbb0032dde5c204e05a0fd8e
Reviewed-on: https://go-review.googlesource.com/23702Reviewed-by: Rick Hudson <rlh@golang.org>

src/runtime/malloc.go

@@ -676,9 +676,7 @@ func mallocgc(size uintptr, typ *_type, needzero bool) unsafe.Pointer {
 	}
 
 	var scanSize uintptr
-	if noscan {
-		heapBitsSetTypeNoScan(uintptr(x))
-	} else {
+	if !noscan {
 		// If allocating a defer+arg block, now that we've picked a malloc size
 		// large enough to hold everything, cut the "asked for" size down to
 		// just the defer header, so that the GC bitmap will record the arg block

src/runtime/mbitmap.go

@@ -45,6 +45,11 @@
 // not checkmarked, and is the dead encoding.
 // These properties must be preserved when modifying the encoding.
 //
+// The bitmap for noscan spans is not maintained. Code must ensure
+// that an object is scannable before consulting its bitmap by
+// checking either the noscan bit in the span or by consulting its
+// type's information.
+//
 // Checkmarks
 //
 // In a concurrent garbage collector, one worries about failing to mark
@@ -1363,13 +1368,6 @@ Phase4:
 	}
 }
 
-// heapBitsSetTypeNoScan marks x as noscan by setting the first word
-// of x in the heap bitmap to scalar/dead.
-func heapBitsSetTypeNoScan(x uintptr) {
-	h := heapBitsForAddr(uintptr(x))
-	*h.bitp &^= (bitPointer | bitMarked) << h.shift
-}
-
 var debugPtrmask struct {
 	lock mutex
 	data *byte