[AArch64] Multiply-Negate operation not lowered to mneg in AArch64

[karthikthecool]

Summary

Hi All,
The below code-

int a,b,c;
void fn() {

c = -(a*b);

}

when compiled on AArch64 bit with -O3 generates the assembly as follows-

fn: @fn
BB#0:
adrp x8, a
adrp x9, b
ldr w8, [x8, :lo12:a]
ldr w9, [x9, :lo12:b]
neg w8, w8
mul w8, w9, w8
adrp x9, c
str w8, [x9, :lo12:c]
ret

the neg and mul operation can be combined to mneg in AArch64. Gcc combines the above to mneg instruction.
Added a patch in llvm to do the same.

Post patch the following assembly is generated-
fn: @fn
BB#0:
adrp x8, a
adrp x9, b
ldr w8, [x8, :lo12:a]
ldr w9, [x9, :lo12:b]
mneg w8, w8, w9
adrp x9, c
str w8, [x9, :lo12:c]
ret

Please let me know if this is good to commit.

Thanks and Regards
Karthik Bhat

[karthikthecool]

Thanks James. Committed as r224706.

lib/Target/AArch64/AArch64InstrInfo.td

def : Pat<(i32 (mul GPR32:$Rn, GPR32:$Rm)),
           (MADDWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
 def : Pat<(i64 (mul GPR64:$Rn, GPR64:$Rm)),
           (MADDXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
 
 def : Pat<(i32 (ineg (mul GPR32:$Rn, GPR32:$Rm))),
           (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
 def : Pat<(i64 (ineg (mul GPR64:$Rn, GPR64:$Rm))),
           (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
+def : Pat<(i32 (mul (ineg GPR32:$Rn), GPR32:$Rm)),
+          (MSUBWrrr GPR32:$Rn, GPR32:$Rm, WZR)>;
+def : Pat<(i64 (mul (ineg GPR64:$Rn), GPR64:$Rm)),
+          (MSUBXrrr GPR64:$Rn, GPR64:$Rm, XZR)>;
 } // AddedComplexity = 7
 
 let AddedComplexity = 5 in {
 def SMADDLrrr : WideMulAccum<0, 0b001, "smaddl", add, sext>;
 def SMSUBLrrr : WideMulAccum<1, 0b001, "smsubl", sub, sext>;
 def UMADDLrrr : WideMulAccum<0, 0b101, "umaddl", add, zext>;
 def UMSUBLrrr : WideMulAccum<1, 0b101, "umsubl", sub, zext>;
 

test/CodeGen/AArch64/dp-3source.ll

 ; CHECK-LABEL: test_umnegl:
   %ext1 = zext i32 %lhs to i64
   %ext2 = zext i32 %rhs to i64
   %prod = mul i64 %ext1, %ext2
   %res = sub i64 0, %prod
 ; CHECK: umnegl {{x[0-9]+}}, {{w[0-9]+}}, {{w[0-9]+}}
   ret i64 %res
 }
+
+@a = common global i32 0, align 4
+@b = common global i32 0, align 4
+@c = common global i32 0, align 4
+
+define void @test_mneg(){
+; CHECK-LABEL: test_mneg:
+  %1 = load i32* @a, align 4
+  %2 = load i32* @b, align 4
+  %3 = sub i32 0, %1
+  %4 = mul i32 %2, %3
+  store i32 %4, i32* @c, align 4
+; CHECK: mneg {{w[0-9]+}}, {{w[0-9]+}}, {{w[0-9]+}}
+  ret void
+}