cmd/compile/internal/ssa: prefer non-indexed stores on amd64

We sometimes generate code like this:

v473 = MOVQconst <uintptr> // constant..
v580 = MOVBstoreidx1 <mem> v1056 v473 v819 v491 // ..only used as an index

Rewrite indexed stores to non-indexed version, where possible.
This allows to eliminate  const->register move, reducing codesize and lowering register pressure.

Change-Id: Id5fed801dffe3f1a80876b8d8bd43775a7c942b2
Reviewed-on: https://go-review.googlesource.com/105295
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/ssa/gen/AMD64.rules

@@ -2416,6 +2416,32 @@
 (BSFQ (ORQconst <t> [1<<8] (MOVBQZX x))) -> (BSFQ (ORQconst <t> [1<<8] x))
 (BSFQ (ORQconst <t> [1<<16] (MOVWQZX x))) -> (BSFQ (ORQconst <t> [1<<16] x))
 
+// Simplify indexed loads/stores
+(MOVBstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVBstore [i+c] {s} p w mem)
+(MOVWstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVWstore [i+c] {s} p w mem)
+(MOVLstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVLstore [i+c] {s} p w mem)
+(MOVQstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVQstore [i+c] {s} p w mem)
+(MOVWstoreidx2 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+2*c) -> (MOVWstore [i+2*c] {s} p w mem)
+(MOVLstoreidx4 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+4*c) -> (MOVLstore [i+4*c] {s} p w mem)
+(MOVLstoreidx8 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+8*c) -> (MOVLstore [i+8*c] {s} p w mem)
+(MOVQstoreidx8 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+8*c) -> (MOVQstore [i+8*c] {s} p w mem)
+(MOVSSstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVSSstore [i+c] {s} p w mem)
+(MOVSSstoreidx4 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+4*c) -> (MOVSSstore [i+4*c] {s} p w mem)
+(MOVSDstoreidx1 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+c) -> (MOVSDstore [i+c] {s} p w mem)
+(MOVSDstoreidx8 [i] {s} p (MOVQconst [c]) w mem) && is32Bit(i+8*c) -> (MOVSDstore [i+8*c] {s} p w mem)
+(MOVBloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVBload [i+c] {s} p mem)
+(MOVWloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVWload [i+c] {s} p mem)
+(MOVLloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVLload [i+c] {s} p mem)
+(MOVQloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVQload [i+c] {s} p mem)
+(MOVWloadidx2 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+2*c) -> (MOVWload [i+2*c] {s} p mem)
+(MOVLloadidx4 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+4*c) -> (MOVLload [i+4*c] {s} p mem)
+(MOVLloadidx8 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+8*c) -> (MOVLload [i+8*c] {s} p mem)
+(MOVQloadidx8 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+8*c) -> (MOVQload [i+8*c] {s} p mem)
+(MOVSSloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVSSload [i+c] {s} p mem)
+(MOVSSloadidx4 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+4*c) -> (MOVSSload [i+4*c] {s} p mem)
+(MOVSDloadidx1 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+c) -> (MOVSDload [i+c] {s} p mem)
+(MOVSDloadidx8 [i] {s} p (MOVQconst [c]) mem) && is32Bit(i+8*c) -> (MOVSDload [i+8*c] {s} p mem)
+
 // Redundant sign/zero extensions
 // Note: see issue 21963. We have to make sure we use the right type on
 // the resulting extension (the outer type, not the inner type).