ARM32 Vector lowering - scalarize select

tests_lit/assembler/arm32/select-vec.ll

Index: tests_lit/assembler/arm32/select-vec.ll
diff --git a/tests_lit/assembler/arm32/select-vec.ll b/tests_lit/assembler/arm32/select-vec.ll
new file mode 100644
index 0000000000000000000000000000000000000000..0a60da4b6ab107a8e9bd2151bc05d4db5775067e
--- /dev/null
+++ b/tests_lit/assembler/arm32/select-vec.ll
@@ -0,0 +1,51 @@
+; Test that we handle select on vectors.
+
+; TODO(eholk): This test will need to be updated once comparison is no
+; longer scalarized.
+
+; REQUIRES: allow_dump
+
+; Compile using standalone assembler.
+; RUN: %p2i --filetype=asm -i %s --target=arm32 --args -O2 \
+; RUN:   | FileCheck %s --check-prefix=ASM
+
+

[Jim Stichnoth, 2016/02/11 15:26:07]

I think these tests generally only have one blank line here.

[Eric Holk, 2016/02/11 19:17:23]

Done.

+
+define internal <4 x i32> @select4i32(<4 x i32> %a, <4 x i32> %b, <4 x i1> %s) {

[Jim Stichnoth, 2016/02/11 15:26:07]

You should probably include tests on all the vector types.

Also, any reasons the function args are listed in a different order from the
select instruction?

[Eric Holk, 2016/02/11 19:17:23]

Okay. How much coverage do we have in the crosstests? I've been using shorter
tests lately because I expect the crosstests to do a better job testing these
instruction soon anyway.
Not really. I can change it.

+; ASM-LABEL:select4i32:
+; DIS-LABEL:00000000 <select4i32>:
+
+entry:
+  %res = select <4 x i1> %s, <4 x i32> %a, <4 x i32> %b
+
+; ASM:	# q3 = def.pseudo
+; ASM-NEXT:	vmov.s8	r0, d4[0]
+; ASM-NEXT:	vmov.32	r1, d0[0]
+; ASM-NEXT:	vmov.32	r2, d2[0]
+; ASM-NEXT:	tst	r0, #1
+; ASM-NEXT:	movne	r2, r1
+; ASM-NEXT:	vmov.32	d6[0], r2
+; ASM-NEXT:	vmov.s8	r0, d4[1]
+; ASM-NEXT:	vmov.32	r1, d0[1]
+; ASM-NEXT:	vmov.32	r2, d2[1]
+; ASM-NEXT:	tst	r0, #1
+; ASM-NEXT:	movne	r2, r1
+; ASM-NEXT:	vmov.32	d6[1], r2
+; ASM-NEXT:	vmov.s8	r0, d5[0]
+; ASM-NEXT:	vmov.32	r1, d1[0]
+; ASM-NEXT:	vmov.32	r2, d3[0]
+; ASM-NEXT:	tst	r0, #1
+; ASM-NEXT:	movne	r2, r1
+; ASM-NEXT:	vmov.32	d7[0], r2
+; ASM-NEXT:	vmov.s8	r0, d5[1]
+; ASM-NEXT:	vmov.32	r1, d1[1]
+; ASM-NEXT:	vmov.32	r2, d3[1]
+; ASM-NEXT:	tst	r0, #1
+; ASM-NEXT:	movne	r2, r1
+; ASM-NEXT:	vmov.32	d7[1], r2
+; ASM-NEXT:	vmov.i32	q0, q3
+; ASM-NEXT:	bx	lr
+
+  ret <4 x i32> %res
+}
+

[Jim Stichnoth, 2016/02/11 15:26:07]

remove trailing blank line