runtime/race: update runtime (v3)

New tsan runtime (v3) built on llvm commit 1784fe0532a6.

The new runtime features:
 - 2x smaller shadow memory (2x of app memory)
 - faster fully vectorized (on x86) race detection
 - small fixed-size vector clocks (512b)
 - fast vectorized vector clock operations
 - unlimited number of alive threads/goroutines

Some random subset of benchmarks:

encoding/json:
CodeEncoder-72           20.3ms ± 6%     11.7ms ± 4%   -42.02%  (p=0.000 n=10+8)
CodeMarshal-72           22.3ms ±11%     12.7ms ±11%   -43.28%  (p=0.000 n=10+10)
CodeDecoder-72           46.1ms ±42%     20.2ms ± 7%   -56.18%  (p=0.000 n=10+10)
CodeUnmarshal-72         16.7ms ±14%     13.2ms ± 9%   -20.93%  (p=0.000 n=10+10)
CodeUnmarshalReuse-72    17.7ms ±17%     12.8ms ± 8%   -27.58%  (p=0.000 n=10+10)

net/http:
ClientServerParallel4-72         914µs ±18%      72µs ± 5%  -92.16%  (p=0.000 n=20+18)
ClientServerParallel64-72       1.77ms ±16%    0.12ms ±29%  -93.43%  (p=0.000 n=20+17)
ClientServerParallelTLS4-72     1.99ms ±33%    0.20ms ± 5%  -89.83%  (p=0.000 n=19+17)
ClientServerParallelTLS64-72    2.72ms ±26%    0.25ms ±16%  -90.96%  (p=0.000 n=20+16)

compress/flate:
Decode/Digits/Huffman/1e4-72        2.53ms ± 3%     1.47ms ± 4%   -41.72%  (p=0.000 n=9+10)
Decode/Digits/Speed/1e4-72          2.59ms ± 5%     1.52ms ± 3%   -41.44%  (p=0.000 n=8+9)
Decode/Digits/Default/1e4-72        2.56ms ± 6%     1.51ms ± 4%   -40.96%  (p=0.000 n=10+10)
Decode/Digits/Compression/1e4-72    2.54ms ± 6%     1.52ms ± 2%   -40.14%  (p=0.000 n=10+9)
Decode/Newton/Huffman/1e4-72        2.65ms ± 3%     1.58ms ± 4%   -40.45%  (p=0.000 n=10+10)
Decode/Newton/Speed/1e4-72          2.16ms ± 9%     1.28ms ± 5%   -40.59%  (p=0.000 n=10+10)
Decode/Newton/Default/1e4-72        2.01ms ± 8%     1.16ms ± 6%   -42.11%  (p=0.000 n=10+10)
Decode/Newton/Compression/1e4-72    1.99ms ± 6%     1.17ms ± 3%   -41.05%  (p=0.000 n=9+10)
Encode/Digits/Huffman/1e4-72        2.05ms ± 2%     0.75ms ± 5%   -63.32%  (p=0.000 n=10+10)
Encode/Digits/Speed/1e4-72          2.89ms ± 2%     1.37ms ± 2%   -52.56%  (p=0.000 n=9+10)
Encode/Digits/Default/1e4-72        7.55ms ± 2%     3.86ms ± 4%   -48.93%  (p=0.000 n=8+10)
Encode/Digits/Compression/1e4-72    7.46ms ± 4%     3.88ms ± 4%   -48.07%  (p=0.000 n=9+9)
Encode/Newton/Huffman/1e4-72        2.20ms ± 4%     0.90ms ± 6%   -59.01%  (p=0.000 n=10+10)
Encode/Newton/Speed/1e4-72          2.62ms ± 2%     1.30ms ± 4%   -50.52%  (p=0.000 n=9+10)
Encode/Newton/Default/1e4-72        7.40ms ± 5%     3.72ms ± 2%   -49.65%  (p=0.000 n=10+9)
Encode/Newton/Compression/1e4-72    7.67ms ± 9%     3.85ms ± 4%   -49.87%  (p=0.000 n=10+10)

encoding/json short tests:
Time     2.34sec ± 6%   1.67sec ±11%  -28.85%  (p=0.000 n=10+10)
Memory     266MB ± 1%     190MB ± 9%  -28.78%  (p=0.000 n=10+10)

compress/flate short tests:
Time      4.69sec ± 4%   2.78sec ± 3%  -40.61%  (p=0.000 n=10+10)
Memory      284MB ± 5%     181MB ± 4%  -36.04%  (p=0.000 n=10+9)

sync short tests:
Time     4.87sec ± 4%  0.87sec ± 6%  -82.21%  (p=0.000 n=10+10)
Memory     147MB ± 3%     99MB ± 8%  -32.47%  (p=0.000 n=10+9)

Fixes #47056
Fixes #38184

Change-Id: I0cf228f2e4cac7778d34d33e46df7c081645f5d5
Reviewed-on: https://go-review.googlesource.com/c/go/+/333529
Run-TryBot: Dmitry Vyukov <dvyukov@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Trust: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>

src/runtime/race/README

@@ -6,7 +6,7 @@ To update the .syso files use golang.org/x/build/cmd/racebuild.
 
 race_darwin_amd64.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 with https://reviews.llvm.org/D114825 applied and Go 7ccbcc90560468937f02609a43cb39a6e13ff797.
 race_freebsd_amd64.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 and Go f62d3202bf9dbb3a00ad2a2c63ff4fa4188c5d3b.
-race_linux_amd64.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 and Go f62d3202bf9dbb3a00ad2a2c63ff4fa4188c5d3b.
+race_linux_amd64.syso built with LLVM 41cb504b7c4b18ac15830107431a0c1eec73a6b2 and Go 851ecea4cc99ab276109493477b2c7e30c253ea8.
 race_linux_ppc64le.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 and Go f62d3202bf9dbb3a00ad2a2c63ff4fa4188c5d3b.
 race_netbsd_amd64.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 and Go f62d3202bf9dbb3a00ad2a2c63ff4fa4188c5d3b.
 race_windows_amd64.syso built with LLVM 89f7ccea6f6488c443655880229c54db1f180153 and Go f62d3202bf9dbb3a00ad2a2c63ff4fa4188c5d3b.