Skip to content

G
golang
Commit 9f291d1f authored by Keith Randall's avatar Keith Randall Committed by Ben Shi
Browse files
Options

cmd/compile: fix rule for combining loads with compares 

Unlike normal load+op opcodes, the load+compare opcode does
not clobber its non-load argument. Allow the load+compare merge
to happen even if the non-load argument is used elsewhere.

Noticed when investigating issue #28417.

Change-Id: Ibc48d1f2e06ae76034c59f453815d263e8ec7288
Reviewed-on: https://go-review.googlesource.com/c/145097Reviewed-by: 's avatarAinar Garipov <gugl.zadolbal@gmail.com>
Reviewed-by: 's avatarBen Shi <powerman1st@163.com>
parent 67a9c0af
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 73 additions and 60 deletions
src/cmd/compile/internal/ssa/gen/386.rules
View file @ 9f291d1f
......@@ -771,14 +771,14 @@
(MOVSDstoreidx8 [c] {sym} ptr (ADDLconst [d] idx) val mem) -> (MOVSDstoreidx8 [int64(int32(c+8*d))] {sym} ptr idx val mem)
// Merge load/store to op
((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> ((ADD|AND|OR|XOR|SUB|MUL)Lload x [off] {sym} ptr mem)
((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLloadidx4 [off] {sym} ptr idx mem)) && canMergeLoad(v, l, x) && clobber(l) ->
((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|AND|OR|XOR|SUB|MUL)Lload x [off] {sym} ptr mem)
((ADD|AND|OR|XOR|SUB|MUL)L x l:(MOVLloadidx4 [off] {sym} ptr idx mem)) && canMergeLoadClobber(v, l, x) && clobber(l) ->
((ADD|AND|OR|XOR|SUB|MUL)Lloadidx4 x [off] {sym} ptr idx mem)
((ADD|SUB|MUL|AND|OR|XOR)Lload [off1] {sym1} val (LEAL4 [off2] {sym2} ptr idx) mem)
&& is32Bit(off1+off2) && canMergeSym(sym1, sym2) ->
((ADD|SUB|MUL|AND|OR|XOR)Lloadidx4 [off1+off2] {mergeSym(sym1,sym2)} val ptr idx mem)
((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && !config.use387 && clobber(l) -> ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && !config.use387 && clobber(l) -> ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && !config.use387 && clobber(l) -> ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && !config.use387 && clobber(l) -> ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem)
(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) -> ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem)
(MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y) && clobber(l) ->
((ADD|SUB|AND|OR|XOR)Lmodify [off] {sym} ptr x mem)
......@@ -1217,8 +1217,8 @@
(MOVSSconst [c]) && config.ctxt.Flag_shared -> (MOVSSconst2 (MOVSSconst1 [c]))
(MOVSDconst [c]) && config.ctxt.Flag_shared -> (MOVSDconst2 (MOVSDconst1 [c]))
(CMP(L|W|B) l:(MOV(L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l, x) && clobber(l) -> (CMP(L|W|B)load {sym} [off] ptr x mem)
(CMP(L|W|B) x l:(MOV(L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (InvertFlags (CMP(L|W|B)load {sym} [off] ptr x mem))
(CMP(L|W|B) l:(MOV(L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l) && clobber(l) -> (CMP(L|W|B)load {sym} [off] ptr x mem)
(CMP(L|W|B) x l:(MOV(L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) -> (InvertFlags (CMP(L|W|B)load {sym} [off] ptr x mem))
(CMP(L|W|B)const l:(MOV(L|W|B)load {sym} [off] ptr mem) [c])
&& l.Uses == 1
......
src/cmd/compile/internal/ssa/gen/AMD64.rules
View file @ 9f291d1f
......@@ -2362,10 +2362,10 @@
// Merge load and op
// TODO: add indexed variants?
((ADD|SUB|AND|OR|XOR)Q x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Qload x [off] {sym} ptr mem)
((ADD|SUB|AND|OR|XOR)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Lload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem)
((ADD|SUB|AND|OR|XOR)Q x l:(MOVQload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Qload x [off] {sym} ptr mem)
((ADD|SUB|AND|OR|XOR)L x l:(MOVLload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|AND|OR|XOR)Lload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SD x l:(MOVSDload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SDload x [off] {sym} ptr mem)
((ADD|SUB|MUL|DIV)SS x l:(MOVSSload [off] {sym} ptr mem)) && canMergeLoadClobber(v, l, x) && clobber(l) -> ((ADD|SUB|MUL|DIV)SSload x [off] {sym} ptr mem)
(MOVLstore {sym} [off] ptr y:((ADD|AND|OR|XOR)Lload x [off] {sym} ptr mem) mem) && y.Uses==1 && clobber(y) -> ((ADD|AND|OR|XOR)Lmodify [off] {sym} ptr x mem)
(MOVLstore {sym} [off] ptr y:((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)L l:(MOVLload [off] {sym} ptr mem) x) mem) && y.Uses==1 && l.Uses==1 && clobber(y) && clobber(l) ->
((ADD|SUB|AND|OR|XOR|BTC|BTR|BTS)Lmodify [off] {sym} ptr x mem)
......@@ -2514,8 +2514,8 @@
// Fold loads into compares
// Note: these may be undone by the flagalloc pass.
(CMP(Q|L|W|B) l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l, x) && clobber(l) -> (CMP(Q|L|W|B)load {sym} [off] ptr x mem)
(CMP(Q|L|W|B) x l:(MOV(Q|L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l, x) && clobber(l) -> (InvertFlags (CMP(Q|L|W|B)load {sym} [off] ptr x mem))
(CMP(Q|L|W|B) l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) x) && canMergeLoad(v, l) && clobber(l) -> (CMP(Q|L|W|B)load {sym} [off] ptr x mem)
(CMP(Q|L|W|B) x l:(MOV(Q|L|W|B)load {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) -> (InvertFlags (CMP(Q|L|W|B)load {sym} [off] ptr x mem))
(CMP(Q|L|W|B)const l:(MOV(Q|L|W|B)load {sym} [off] ptr mem) [c])
&& l.Uses == 1
......
src/cmd/compile/internal/ssa/gen/S390X.rules
View file @ 9f291d1f
......@@ -1125,71 +1125,71 @@
// Exclude global data (SB) because these instructions cannot handle relative addresses.
// TODO(mundaym): use LARL in the assembler to handle SB?
// TODO(mundaym): indexed versions of these?
(ADD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDload <t> [off] {sym} x ptr mem)
(ADD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDload <t> [off] {sym} x ptr mem)
(ADDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDWload <t> [off] {sym} x ptr mem)
(ADDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDWload <t> [off] {sym} x ptr mem)
(ADDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDWload <t> [off] {sym} x ptr mem)
(ADDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ADDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ADDWload <t> [off] {sym} x ptr mem)
(MULLD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLD <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLDload <t> [off] {sym} x ptr mem)
(MULLD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLD <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLDload <t> [off] {sym} x ptr mem)
(MULLW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLWload <t> [off] {sym} x ptr mem)
(MULLW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLWload <t> [off] {sym} x ptr mem)
(MULLW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLWload <t> [off] {sym} x ptr mem)
(MULLW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(MULLW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (MULLWload <t> [off] {sym} x ptr mem)
(SUB <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(SUB <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (SUBload <t> [off] {sym} x ptr mem)
(SUBW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(SUBW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (SUBWload <t> [off] {sym} x ptr mem)
(SUBW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(SUBW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (SUBWload <t> [off] {sym} x ptr mem)
(AND <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(AND <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDload <t> [off] {sym} x ptr mem)
(AND <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(AND <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDload <t> [off] {sym} x ptr mem)
(ANDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ANDW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDWload <t> [off] {sym} x ptr mem)
(ANDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ANDW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDWload <t> [off] {sym} x ptr mem)
(ANDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ANDW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDWload <t> [off] {sym} x ptr mem)
(ANDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ANDW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ANDWload <t> [off] {sym} x ptr mem)
(OR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(OR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORload <t> [off] {sym} x ptr mem)
(OR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(OR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORload <t> [off] {sym} x ptr mem)
(ORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORWload <t> [off] {sym} x ptr mem)
(ORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORWload <t> [off] {sym} x ptr mem)
(ORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORWload <t> [off] {sym} x ptr mem)
(ORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(ORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (ORWload <t> [off] {sym} x ptr mem)
(XOR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XOR <t> x g:(MOVDload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORload <t> [off] {sym} x ptr mem)
(XOR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XOR <t> g:(MOVDload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORload <t> [off] {sym} x ptr mem)
(XORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XORW <t> x g:(MOVWload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORWload <t> [off] {sym} x ptr mem)
(XORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XORW <t> g:(MOVWload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORWload <t> [off] {sym} x ptr mem)
(XORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XORW <t> x g:(MOVWZload [off] {sym} ptr mem)) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORWload <t> [off] {sym} x ptr mem)
(XORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoad(v, g, x) && clobber(g)
(XORW <t> g:(MOVWZload [off] {sym} ptr mem) x) && ptr.Op != OpSB && is20Bit(off) && canMergeLoadClobber(v, g, x) && clobber(g)
-> (XORWload <t> [off] {sym} x ptr mem)
// Combine constant stores into larger (unaligned) stores.
......
src/cmd/compile/internal/ssa/rewrite.go
View file @ 9f291d1f
......@@ -177,23 +177,11 @@ func canMergeSym(x, y interface{}) bool {
return x == nil || y == nil
}
// canMergeLoad reports whether the load can be merged into target without
// canMergeLoadClobber reports whether the load can be merged into target without
// invalidating the schedule.
// It also checks that the other non-load argument x is something we
// are ok with clobbering (all our current load+op instructions clobber
// their input register).
func canMergeLoad(target, load, x *Value) bool {
if target.Block.ID != load.Block.ID {
// If the load is in a different block do not merge it.
return false
}
// We can't merge the load into the target if the load
// has more than one use.
if load.Uses != 1 {
return false
}
// are ok with clobbering.
func canMergeLoadClobber(target, load, x *Value) bool {
// The register containing x is going to get clobbered.
// Don't merge if we still need the value of x.
// We don't have liveness information here, but we can
......@@ -208,6 +196,22 @@ func canMergeLoad(target, load, x *Value) bool {
if loopnest.depth(target.Block.ID) > loopnest.depth(x.Block.ID) {
return false
}
return canMergeLoad(target, load)
}
// canMergeLoad reports whether the load can be merged into target without
// invalidating the schedule.
func canMergeLoad(target, load *Value) bool {
if target.Block.ID != load.Block.ID {
// If the load is in a different block do not merge it.
return false
}
// We can't merge the load into the target if the load
// has more than one use.
if load.Uses != 1 {
return false
}
mem := load.MemoryArg()
......
src/cmd/compile/internal/ssa/rewrite386.go
View file @ 9f291d1f
This diff is collapsed.
test/codegen/memops.go
View file @ 9f291d1f
......@@ -84,3 +84,12 @@ func compMem2() int {
}
return 0
}
func compMem3(x, y *int) (int, bool) {
// We can do comparisons of a register with memory even if
// the register is used subsequently.
r := *x
// amd64:`CMPQ\t\(`
// 386:`CMPL\t\(`
return r, r < *y
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment