Subzero. X86. Lowers shufflevector using xmm instructions.

tests_lit/llvm2ice_tests/vector-shuffle.ll

Index: tests_lit/llvm2ice_tests/vector-shuffle.ll
diff --git a/tests_lit/llvm2ice_tests/vector-shuffle.ll b/tests_lit/llvm2ice_tests/vector-shuffle.ll
new file mode 100644
index 0000000000000000000000000000000000000000..016342f3f68290fdf2a7ad64a2d1a4c8f14dd18c
--- /dev/null
+++ b/tests_lit/llvm2ice_tests/vector-shuffle.ll
@@ -0,0 +1,53 @@
+; Some shufflevector optimized lowering. This list is by no means exhaustive.

[Jim Stichnoth, 2016/04/25 21:23:45]

reflow to 80-col

[John, 2016/04/25 22:38:39]

Done.

+; It is only a **basic** smoke test. the vector_ops crosstest has a broader
+; range of test cases.
+
+; RUN: %p2i -i %s --target=x8632 --filetype=obj --disassemble -a -O2 \
+; RUN:     --allow-externally-defined-symbols | FileCheck %s --check-prefix=X86
+
+declare void @useV4I32(<4 x i32> %t);
+
+define internal void @shuffleV4I32(<4 x i32> %a, <4 x i32> %b) {
+; X86-LABEL: shuffleV4I32
+  %a_0 = extractelement <4 x i32> %a, i32 0
+  %a_1 = extractelement <4 x i32> %a, i32 1
+  %a_2 = extractelement <4 x i32> %a, i32 2
+  %a_3 = extractelement <4 x i32> %a, i32 3
+
+  %b_0 = extractelement <4 x i32> %b, i32 0
+  %b_1 = extractelement <4 x i32> %b, i32 1
+  %b_2 = extractelement <4 x i32> %b, i32 2
+  %b_3 = extractelement <4 x i32> %b, i32 3
+
+  %t0_0 = insertelement <4 x i32> undef, i32 %a_0, i32 0
+  %t0_1 = insertelement <4 x i32> %t0_0, i32 %b_0, i32 1
+  %t0_2 = insertelement <4 x i32> %t0_1, i32 %a_1, i32 2
+  %t0   = insertelement <4 x i32> %t0_2, i32 %b_1, i32 3
+; X86: punpckldq {{.*}}
+
+  call void @useV4I32(<4 x i32> %t0)
+; X86: call
+
+  %t1_0 = insertelement <4 x i32> undef, i32 %a_0, i32 0
+  %t1_1 = insertelement <4 x i32> %t1_0, i32 %b_1, i32 1
+  %t1_2 = insertelement <4 x i32> %t1_1, i32 %b_1, i32 2
+  %t1   = insertelement <4 x i32> %t1_2, i32 %a_0, i32 3
+; X86: shufps [[T:xmm[0-9]+]],{{.*}},0x10
+; X86: pshufd {{.*}},[[T]],0x28
+
+  call void @useV4I32(<4 x i32> %t1)
+; X86: call
+
+  %t2_0 = insertelement <4 x i32> undef, i32 %a_0, i32 0
+  %t2_1 = insertelement <4 x i32> %t2_0, i32 %b_3, i32 1
+  %t2_2 = insertelement <4 x i32> %t2_1, i32 %a_2, i32 2
+  %t2   = insertelement <4 x i32> %t2_2, i32 %b_2, i32 3
+; X86: shufps {{.*}},0x30
+; X86: shufps {{.*}},0x22
+; X86: shufps {{.*}},0x88
+
+  call void @useV4I32(<4 x i32> %t2)
+; X86: call
+
+  ret void
+}