1 ; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 -O0 < %s | FileCheck %s
3 ; Check that at -O0, the backend doesn't attempt to canonicalize a vector load
4 ; used by an INSERTPS into a scalar load plus scalar_to_vector.
6 ; In order to fold a load into the memory operand of an INSERTPSrm, the backend
7 ; tries to canonicalize a vector load in input to an INSERTPS node into a
8 ; scalar load plus scalar_to_vector. This would allow ISel to match the
9 ; INSERTPSrm variant rather than a load plus INSERTPSrr.
11 ; However, ISel can only select an INSERTPSrm if folding a load into the operand
12 ; of an insertps is considered to be profitable.
14 ; In the example below:
16 ; __m128 test(__m128 a, __m128 *b) {
17 ; __m128 c = _mm_insert_ps(a, *b, 1 << 6);
21 ; At -O0, the backend would attempt to canonicalize the load to 'b' into
22 ; a scalar load in the hope of matching an INSERTPSrm.
23 ; However, ISel would fail to recognize an INSERTPSrm since load folding is
24 ; always considered unprofitable at -O0. This would leave the insertps mask
25 ; in an invalid state.
27 ; The problem with the canonicalization rule performed by the backend is that
28 ; it assumes ISel to always be able to match an INSERTPSrm. This assumption is
29 ; not always correct at -O0. In this example, FastISel fails to lower the
30 ; arguments needed by the entry block. This is enough to enable the DAGCombiner
31 ; and eventually trigger the canonicalization on the INSERTPS node.
33 ; This test checks that the vector load in input to the insertps is not
34 ; canonicalized into a scalar load plus scalar_to_vector (a movss).
36 define <4 x float> @test(<4 x float> %a, ptr %b) {
38 ; CHECK: movaps (%rdi), [[REG:%[a-z0-9]+]]
40 ; CHECK: insertps $64, [[REG]],
43 %0 = load <4 x float>, ptr %b, align 16
44 %1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %0, i32 64)
45 %2 = alloca <4 x float>, align 16
46 store <4 x float> %1, ptr %2, align 16
47 %3 = load <4 x float>, ptr %2, align 16
52 declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i32)