Go: cmd/compile: regression in 69a7c15

[OneOfOne]

Please answer these questions before submitting your issue. Thanks!

1. What version of Go are you using (go version)?
   go version devel +5cdb3d0 2016-03-22 20:35:53 +0000 linux/amd64
2. What operating system and processor architecture are you using (go env)?
   Linux amd64
3. What did you do?

go get -u github.com/OneOfOne/xxhash
go test -bench=XXChecksum64\$ -benchtime 3s -tags safe github.com/OneOfOne/xxhash

1. What did you expect to see?
   BenchmarkXXChecksum64-32 20000000 291 ns/op
2. What did you see instead?
   BenchmarkXXChecksum64-8 5000000 870 ns/op

I ran a bisect and the offending commit is 69a7c152a72c713032498bfbc6ec7c41d84a4b63.

[randall77]

This looks like 69a7c15 broke the combine-8-byte-loads-into-a-quadword-load optimization. That's what the slowdown is coming from. I'll see if I can figure out why.

[gopherbot]

CL https://golang.org/cl/21004 mentions this issue.

[gopherbot]

CL https://golang.org/cl/21005 mentions this issue.