When using AVX vectors double x[2], a[4], b[4], c[5]; void foo () { a[0] = c[0]; a[1] = c[1]; a[2] = c[0]; a[3] = c[1]; b[0] = c[2]; b[1] = c[3]; b[2] = c[2]; b[3] = c[3]; x[0] = c[4]; x[1] = c[4]; } only vectorizes the x[] stores since the overall SLP analysis succeeds with V4DFmode but only parts of the opportunities are finally vectorized and thus SLP vectorization with V2DFmode isn't even tried. This is the issue that vector mode iteration works on wrong granularity. /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:17:8: note: === vect_slp_analyze_instance_dependence === /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:9:8: note: === vect_slp_analyze_instance_alignment === /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:9:8: note: removing SLP instance operations starting from: a[0] = _1; /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:13:8: note: === vect_slp_analyze_instance_alignment === /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:13:8: note: removing SLP instance operations starting from: b[0] = _3; /home/rguenther/src/gcc3/gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c:13:8: note: === vect_slp_analyze_operations === and the failure is because of the now late performed /* Analyze alignment of DRs of stmts in NODE. */ static bool vect_slp_analyze_node_alignment (vec_info *vinfo, slp_tree node) { ... /* We need to commit to a vector type for the group now. */ if (is_a <bb_vec_info> (vinfo) && !vect_update_shared_vectype (first_stmt_info, SLP_TREE_VECTYPE (node))) return false; because the other SLP instance (with x[] stores) set the vector type to V2DF while this one wants V4DF.
A similar case is gcc.dg/vect/bb-slp-pr65935.c where based on luck we vectorize either a large leading AVX chain or a single SSE chain.
So a simpler testcase is the following (but hinting at the possibly not generic enough solution to split the load group): double a[6], b[6]; void foo() { a[0] = b[0]; a[1] = b[1]; a[2] = b[2]; a[3] = b[3]; a[4] = b[4]; a[5] = b[5]; } produces with SSE: movapd b(%rip), %xmm0 movapd b+16(%rip), %xmm1 movapd b+32(%rip), %xmm2 movaps %xmm0, a(%rip) movaps %xmm1, a+16(%rip) movaps %xmm2, a+32(%rip) and with AVX: vmovsd b+32(%rip), %xmm0 vmovapd b(%rip), %ymm1 vmovsd %xmm0, a+32(%rip) vmovsd b+40(%rip), %xmm0 vmovapd %ymm1, a(%rip) vmovsd %xmm0, a+40(%rip) while we'd like to see sth like vmovapd b(%rip), %ymm1 vmovapd %ymm1, a(%rip) movapd b+32(%rip), %xmm2 movaps %xmm2, a+32(%rip)
Confirmed.
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>: https://gcc.gnu.org/g:6390c5047adb75960f86d56582e6322aaa4d9281 commit r12-3893-g6390c5047adb75960f86d56582e6322aaa4d9281 Author: Richard Biener <rguenther@suse.de> Date: Wed Nov 18 09:36:57 2020 +0100 Allow different vector types for stmt groups This allows vectorization (in practice non-loop vectorization) to have a stmt participate in different vector type vectorizations. It allows us to remove vect_update_shared_vectype and replace it by pushing/popping STMT_VINFO_VECTYPE from SLP_TREE_VECTYPE around vect_analyze_stmt and vect_transform_stmt. For data-ref the situation is a bit more complicated since we analyze alignment info with a specific vector type in mind which doesn't play well when that changes. So the bulk of the change is passing down the actual vector type used for a vectorized access to the various accessors of alignment info, first and foremost dr_misalignment but also aligned_access_p, known_alignment_for_access_p, vect_known_alignment_in_bytes and vect_supportable_dr_alignment. I took the liberty to replace ALL_CAPS macro accessors with the lower-case function invocations. The actual changes to the behavior are in dr_misalignment which now is the place factoring in the negative step adjustment as well as handling alignment queries for a vector type with bigger alignment requirements than what we can (or have) analyze(d). vect_slp_analyze_node_alignment makes use of this and upon receiving a vector type with a bigger alingment desire re-analyzes the DR with respect to it but keeps an older more precise result if possible. In this context it might be possible to do the analysis just once but instead of analyzing with respect to a specific desired alignment look for the biggest alignment we can compute a not unknown alignment. The ChangeLog includes the functional changes but not the bulk due to the alignment accessor API changes - I hope that's something good. 2021-09-17 Richard Biener <rguenther@suse.de> PR tree-optimization/97351 PR tree-optimization/97352 PR tree-optimization/82426 * tree-vectorizer.h (dr_misalignment): Add vector type argument. (aligned_access_p): Likewise. (known_alignment_for_access_p): Likewise. (vect_supportable_dr_alignment): Likewise. (vect_known_alignment_in_bytes): Likewise. Refactor. (DR_MISALIGNMENT): Remove. (vect_update_shared_vectype): Likewise. * tree-vect-data-refs.c (dr_misalignment): Refactor, handle a vector type with larger alignment requirement and apply the negative step adjustment here. (vect_calculate_target_alignment): Remove. (vect_compute_data_ref_alignment): Get explicit vector type argument, do not apply a negative step alignment adjustment here. (vect_slp_analyze_node_alignment): Re-analyze alignment when we re-visit the DR with a bigger desired alignment but keep more precise results from smaller alignments. * tree-vect-slp.c (vect_update_shared_vectype): Remove. (vect_slp_analyze_node_operations_1): Do not update the shared vector type on stmts. * tree-vect-stmts.c (vect_analyze_stmt): Push/pop the vector type of an SLP node to the representative stmt-info. (vect_transform_stmt): Likewise. * gcc.target/i386/vect-pr82426.c: New testcase. * gcc.target/i386/vect-pr97352.c: Likewise.
Fixed for GCC 12.