cmd/compile: expand calls cleanup

Convert expand calls into a smaller number of focused
recursive rewrites, and rely on an enhanced version of
"decompose" to clean up afterwards.

Debugging information seems to emerge intact.

Change-Id: Ic46da4207e3a4da5c8e2c47b637b0e35abbe56bb
Reviewed-on: https://go-review.googlesource.com/c/go/+/507295
Run-TryBot: David Chase <[email protected]>
Reviewed-by: Cherry Mui <[email protected]>
Reviewed-by: Keith Randall <[email protected]>
TryBot-Result: Gopher Robot <[email protected]>

Author: dr2chase

src/cmd/compile/internal/ssa/_gen/dec.rules

@@ -91,3 +91,109 @@
     (OffPtr <typ.BytePtrPtr> [config.PtrSize] dst)
     data
     (Store {typ.Uintptr} dst itab mem))
+
+// Helpers for expand calls
+// Some of these are copied from generic.rules
+
+(IMake _typ (StructMake1 val)) => (IMake _typ val)
+(StructSelect [0] (IData x)) => (IData x)
+
+(StructSelect (StructMake1 x)) => x
+(StructSelect [0] (StructMake2 x _)) => x
+(StructSelect [1] (StructMake2 _ x)) => x
+(StructSelect [0] (StructMake3 x _ _)) => x
+(StructSelect [1] (StructMake3 _ x _)) => x
+(StructSelect [2] (StructMake3 _ _ x)) => x
+(StructSelect [0] (StructMake4 x _ _ _)) => x
+(StructSelect [1] (StructMake4 _ x _ _)) => x
+(StructSelect [2] (StructMake4 _ _ x _)) => x
+(StructSelect [3] (StructMake4 _ _ _ x)) => x
+
+// Special case coming from immediate interface rewriting
+// Typical case: (StructSelect [0] (IData (IMake typ dat)) rewrites to (StructSelect [0] dat)
+// but because the interface is immediate, the type of "IData" is a one-element struct containing
+// a pointer that is not the pointer type of dat (can be a *uint8).
+// More annoying case: (ArraySelect[0] (StructSelect[0] isAPtr))
+// There, result of the StructSelect is an Array (not a pointer) and
+// the pre-rewrite input to the ArraySelect is a struct, not a pointer.
+(StructSelect [0] x) && x.Type.IsPtr()  => x
+(ArraySelect [0] x) && x.Type.IsPtr()  => x
+
+// These, too.  Bits is bits.
+(ArrayMake1 x) && x.Type.IsPtr() => x
+(StructMake1 x) && x.Type.IsPtr() => x
+
+(Store dst (StructMake1 <t> f0) mem) =>
+  (Store {t.FieldType(0)} (OffPtr <t.FieldType(0).PtrTo()> [0] dst) f0 mem)
+(Store dst (StructMake2 <t> f0 f1) mem) =>
+  (Store {t.FieldType(1)}
+    (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
+    f1
+    (Store {t.FieldType(0)}
+      (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
+        f0 mem))
+(Store dst (StructMake3 <t> f0 f1 f2) mem) =>
+  (Store {t.FieldType(2)}
+    (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] dst)
+    f2
+    (Store {t.FieldType(1)}
+      (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
+      f1
+      (Store {t.FieldType(0)}
+        (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
+          f0 mem)))
+(Store dst (StructMake4 <t> f0 f1 f2 f3) mem) =>
+  (Store {t.FieldType(3)}
+    (OffPtr <t.FieldType(3).PtrTo()> [t.FieldOff(3)] dst)
+    f3
+    (Store {t.FieldType(2)}
+      (OffPtr <t.FieldType(2).PtrTo()> [t.FieldOff(2)] dst)
+      f2
+      (Store {t.FieldType(1)}
+        (OffPtr <t.FieldType(1).PtrTo()> [t.FieldOff(1)] dst)
+        f1
+        (Store {t.FieldType(0)}
+          (OffPtr <t.FieldType(0).PtrTo()> [0] dst)
+            f0 mem))))
+
+(ArraySelect (ArrayMake1 x)) => x
+(ArraySelect [0] (IData x)) => (IData x)
+
+(Store dst (ArrayMake1 e) mem) => (Store {e.Type} dst e mem)
+
+// NOTE removed must-not-be-SSA condition.
+(ArraySelect [i] x:(Load <t> ptr mem)) =>
+  @x.Block (Load <v.Type> (OffPtr <v.Type.PtrTo()> [t.Elem().Size()*i] ptr) mem)
+
+(StringPtr x:(Load <t> ptr mem)) && t.IsString() => @x.Block (Load <typ.BytePtr> ptr mem)
+(StringLen x:(Load <t> ptr mem)) && t.IsString() => @x.Block (Load <typ.Int>
+      (OffPtr <typ.IntPtr> [config.PtrSize] ptr)
+      mem)
+
+// NOTE removed must-not-be-SSA condition.
+(StructSelect [i] x:(Load <t> ptr mem)) =>
+  @x.Block (Load <v.Type> (OffPtr <v.Type.PtrTo()> [t.FieldOff(int(i))] ptr) mem)
+
+(ITab x:(Load <t> ptr mem)) && t.IsInterface() => @x.Block (Load <typ.Uintptr> ptr mem)
+
+(IData x:(Load <t> ptr mem)) && t.IsInterface() => @x.Block (Load <typ.BytePtr>
+      (OffPtr <typ.BytePtrPtr> [config.PtrSize] ptr)
+      mem)
+
+(SlicePtr x:(Load <t> ptr mem)) && t.IsSlice() => @x.Block (Load <t.Elem().PtrTo()> ptr mem)
+(SliceLen x:(Load <t> ptr mem)) && t.IsSlice() => @x.Block (Load <typ.Int>
+      (OffPtr <typ.IntPtr> [config.PtrSize] ptr)
+      mem)
+(SliceCap x:(Load <t> ptr mem)) && t.IsSlice() => @x.Block (Load <typ.Int>
+      (OffPtr <typ.IntPtr> [2*config.PtrSize] ptr)
+      mem)
+
+(ComplexReal x:(Load <t> ptr mem)) && t.IsComplex() && t.Size() == 8 => @x.Block (Load <typ.Float32> ptr mem)
+(ComplexImag x:(Load <t> ptr mem)) && t.IsComplex() && t.Size() == 8 => @x.Block (Load <typ.Float32>
+      (OffPtr <typ.Float32Ptr> [4] ptr)
+      mem)
+
+(ComplexReal x:(Load <t> ptr mem)) && t.IsComplex() && t.Size() == 16 => @x.Block (Load <typ.Float64> ptr mem)
+(ComplexImag x:(Load <t> ptr mem)) && t.IsComplex() && t.Size() == 16 => @x.Block (Load <typ.Float64>
+      (OffPtr <typ.Float64Ptr> [8] ptr)
+      mem)

src/cmd/compile/internal/ssa/compile.go

@@ -472,8 +472,8 @@ var passes = [...]pass{
 	{name: "nilcheckelim", fn: nilcheckelim},
 	{name: "prove", fn: prove},
 	{name: "early fuse", fn: fuseEarly},
-	{name: "decompose builtin", fn: decomposeBuiltIn, required: true},
 	{name: "expand calls", fn: expandCalls, required: true},
+	{name: "decompose builtin", fn: postExpandCallsDecompose, required: true},
 	{name: "softfloat", fn: softfloat, required: true},
 	{name: "late opt", fn: opt, required: true}, // TODO: split required rules and optimizing rules
 	{name: "dead auto elim", fn: elimDeadAutosGeneric},
@@ -547,6 +547,8 @@ var passOrder = [...]constraint{
 	{"generic cse", "tighten"},
 	// checkbce needs the values removed
 	{"generic deadcode", "check bce"},
+	// decompose builtin now also cleans up after expand calls
+	{"expand calls", "decompose builtin"},
 	// don't run optimization pass until we've decomposed builtin objects
 	{"decompose builtin", "late opt"},
 	// decompose builtin is the last pass that may introduce new float ops, so run softfloat after it