cmd/compile: combine x*n - y*n into (x-y)*n

Do the similar thing to CL 55143 to reduce IMUL.

Change-Id: I1bd38f618058e3cd74fac181f003610ea13f2294
Reviewed-on: https://go-review.googlesource.com/56252
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/gc/asm_test.go

@@ -831,6 +831,20 @@ var linuxAMD64Tests = []*asmTest{
 		}`,
 		pos: []string{"\tADDQ\t[$]19", "\tIMULQ"}, // (a+19)*n
 	},
+	{
+		`
+		func mul4(n int) int {
+			return 23*n - 9*n
+		}`,
+		[]string{"\tIMULQ\t[$]14"}, // 14*n
+	},
+	{
+		`
+		func mul5(a, n int) int {
+			return a*n - 19*n
+		}`,
+		[]string{"\tADDQ\t[$]-19", "\tIMULQ"}, // (a-19)*n
+	},
 
 	// see issue 19595.
 	// We want to merge load+op in f58, but not in f59.
@@ -1150,6 +1164,20 @@ var linux386Tests = []*asmTest{
 		`,
 		pos: []string{"TEXT\t.*, [$]0-4"},
 	},
+	{
+		`
+		func mul3(n int) int {
+			return 23*n - 9*n
+		}`,
+		[]string{"\tIMULL\t[$]14"}, // 14*n
+	},
+	{
+		`
+		func mul4(a, n int) int {
+			return n*a - a*19
+		}`,
+		[]string{"\tADDL\t[$]-19", "\tIMULL"}, // (n-19)*a
+	},
 }
 
 var linuxS390XTests = []*asmTest{

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

@@ -328,6 +328,12 @@
 (Add16 <t> (Mul16 x y) (Mul16 x z)) -> (Mul16 x (Add16 <t> y z))
 (Add8  <t> (Mul8  x y) (Mul8  x z)) -> (Mul8  x (Add8  <t> y z))
 
+// Rewrite x*y - x*z  to  x*(y-z)
+(Sub64 <t> (Mul64 x y) (Mul64 x z)) -> (Mul64 x (Sub64 <t> y z))
+(Sub32 <t> (Mul32 x y) (Mul32 x z)) -> (Mul32 x (Sub32 <t> y z))
+(Sub16 <t> (Mul16 x y) (Mul16 x z)) -> (Mul16 x (Sub16 <t> y z))
+(Sub8  <t> (Mul8  x y) (Mul8  x z)) -> (Mul8  x (Sub8  <t> y z))
+
 // rewrite shifts of 8/16/32 bit consts into 64 bit consts to reduce
 // the number of the other rewrite rules for const shifts
 (Lsh64x32  <t> x (Const32 [c])) -> (Lsh64x64  x (Const64 <t> [int64(uint32(c))]))

test/mergemul.go

@@ -24,7 +24,7 @@ import "fmt"
 //   if a8 != b8 {
 // 	   // print error msg and panic
 //   }
-func makeMergeTest(m1, m2, k int, size string) string {
+func makeMergeAddTest(m1, m2, k int, size string) string {
 
 	model := "    a" + size + ", b" + size
 	model += fmt.Sprintf(" = %%d*n%s + %%d*(n%s+%%d), (%%d+%%d)*n%s + (%%d*%%d)", size, size, size)
@@ -32,7 +32,39 @@ func makeMergeTest(m1, m2, k int, size string) string {
 	test := fmt.Sprintf(model, m1, m2, k, m1, m2, m2, k)
 	test += fmt.Sprintf(`
     if a%s != b%s {
-        fmt.Printf("MergeTest(%d, %d, %d, %s) failed\n")
+        fmt.Printf("MergeAddTest(%d, %d, %d, %s) failed\n")
+        fmt.Printf("%%d != %%d\n", a%s, b%s)
+        panic("FAIL")
+    }
+`, size, size, m1, m2, k, size, size, size)
+	return test + "\n"
+}
+
+// Check that expressions like (c*n - d*(n+k)) get correctly merged by
+// the compiler into (c-d)*n - d*k (with c-d and d*k computed at
+// compile time).
+//
+// The merging is performed by a combination of the multiplication
+// merge rules
+//  (c*n - d*n) -> (c-d)*n
+// and the distributive multiplication rules
+//  c * (d-x)  ->  c*d - c*x
+
+// Generate a MergeTest that looks like this:
+//
+//   a8, b8 = m1*n8 - m2*(n8+k), (m1-m2)*n8 - m2*k
+//   if a8 != b8 {
+// 	   // print error msg and panic
+//   }
+func makeMergeSubTest(m1, m2, k int, size string) string {
+
+	model := "    a" + size + ", b" + size
+	model += fmt.Sprintf(" = %%d*n%s - %%d*(n%s+%%d), (%%d-%%d)*n%s - (%%d*%%d)", size, size, size)
+
+	test := fmt.Sprintf(model, m1, m2, k, m1, m2, m2, k)
+	test += fmt.Sprintf(`
+    if a%s != b%s {
+        fmt.Printf("MergeSubTest(%d, %d, %d, %s) failed\n")
         fmt.Printf("%%d != %%d\n", a%s, b%s)
         panic("FAIL")
     }
@@ -42,10 +74,14 @@ func makeMergeTest(m1, m2, k int, size string) string {
 
 func makeAllSizes(m1, m2, k int) string {
 	var tests string
-	tests += makeMergeTest(m1, m2, k, "8")
-	tests += makeMergeTest(m1, m2, k, "16")
-	tests += makeMergeTest(m1, m2, k, "32")
-	tests += makeMergeTest(m1, m2, k, "64")
+	tests += makeMergeAddTest(m1, m2, k, "8")
+	tests += makeMergeAddTest(m1, m2, k, "16")
+	tests += makeMergeAddTest(m1, m2, k, "32")
+	tests += makeMergeAddTest(m1, m2, k, "64")
+	tests += makeMergeSubTest(m1, m2, k, "8")
+	tests += makeMergeSubTest(m1, m2, k, "16")
+	tests += makeMergeSubTest(m1, m2, k, "32")
+	tests += makeMergeSubTest(m1, m2, k, "64")
 	tests += "\n"
 	return tests
 }