https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120234
--- Comment #3 from Haochen Jiang <haochen.jiang at intel dot com> --- The FAIL is caused by the high vec_construct cost after the change. Under -m32, for pr111023 we have: /export/users/haochenj/src/gcc/master/gcc/testsuite/gcc.target/i386/pr111023.c:14:10: note: Cost model analysis: _2 1 times scalar_store costs 16 in body _4 1 times scalar_store costs 16 in body _6 1 times scalar_store costs 16 in body _8 1 times scalar_store costs 16 in body _10 1 times scalar_store costs 16 in body _12 1 times scalar_store costs 16 in body _14 1 times scalar_store costs 16 in body _16 1 times scalar_store costs 16 in body _2 1 times vector_store costs 16 in body node 0x71fd320 1 times vec_construct costs 124 in prologue /export/users/haochenj/src/gcc/master/gcc/testsuite/gcc.target/i386/pr111023.c:14:10: note: Cost model analysis for part in loop 0: Vector cost: 156 Scalar cost: 128 Before the commit, it is: /export/users/haochenj/src/gcc/tmp/gcc/testsuite/gcc.target/i386/pr111023.c:14:10: note: Cost model analysis: _2 1 times scalar_store costs 16 in body _4 1 times scalar_store costs 16 in body _6 1 times scalar_store costs 16 in body _8 1 times scalar_store costs 16 in body _10 1 times scalar_store costs 16 in body _12 1 times scalar_store costs 16 in body _14 1 times scalar_store costs 16 in body _16 1 times scalar_store costs 16 in body _2 1 times vector_store costs 16 in body node 0x78392b0 1 times vec_construct costs 76 in prologue /export/users/haochenj/src/gcc/tmp/gcc/testsuite/gcc.target/i386/pr111023.c:14:10: note: Cost model analysis for part in loop 0: Vector cost: 108 Scalar cost: 128 It leads to the refuse to vectorize.
