1; RUN: llc < %s -march=x86 -mattr=sse41 -stack-alignment=16 -join-physregs > %t 2; RUN: grep pmul %t | count 12 3; RUN: grep mov %t | count 11 4 5; The f() arguments in %xmm0 and %xmm1 cause an extra movdqa without -join-physregs. 6 7define <4 x i32> @a(<4 x i32> %i) nounwind { 8 %A = mul <4 x i32> %i, < i32 117, i32 117, i32 117, i32 117 > 9 ret <4 x i32> %A 10} 11define <2 x i64> @b(<2 x i64> %i) nounwind { 12 %A = mul <2 x i64> %i, < i64 117, i64 117 > 13 ret <2 x i64> %A 14} 15define <4 x i32> @c(<4 x i32> %i, <4 x i32> %j) nounwind { 16 %A = mul <4 x i32> %i, %j 17 ret <4 x i32> %A 18} 19define <2 x i64> @d(<2 x i64> %i, <2 x i64> %j) nounwind { 20 %A = mul <2 x i64> %i, %j 21 ret <2 x i64> %A 22} 23; Use a call to force spills. 24declare void @foo() 25define <4 x i32> @e(<4 x i32> %i, <4 x i32> %j) nounwind { 26 call void @foo() 27 %A = mul <4 x i32> %i, %j 28 ret <4 x i32> %A 29} 30define <2 x i64> @f(<2 x i64> %i, <2 x i64> %j) nounwind { 31 call void @foo() 32 %A = mul <2 x i64> %i, %j 33 ret <2 x i64> %A 34} 35