gcc/adjust-vector-cost-and-move-EXTRACT_LAST_REDUCTION-costing.patch
eastb233 01e0ec8ea6 Upload GCC feature and bugfix patches.
- avoid-cycling-on-vertain-subreg-reloads.patch: Add patch source comment
- change-gcc-BASE-VER.patch: Likewise
- dont-generate-IF_THEN_ELSE.patch: Likewise
- fix-ICE-in-compute_live_loop_exits.patch: Likewise
- fix-ICE-in-eliminate_stmt.patch: Likewise
- fix-ICE-in-vect_create_epilog_for_reduction.patch: Likewise
- fix-ICE-in-vect_stmt_to_vectorize.patch: Likewise
- fix-ICE-in-verify_ssa.patch: Likewise
- fix-ICE-when-vectorizing-nested-cycles.patch: Likewise
- fix-cost-of-plus.patch: Likewise
- ipa-const-prop-self-recursion-bugfix.patch: Likewise
- simplify-removing-subregs.patch: Likewise
- medium-code-mode.patch: Bugfix
- fix-when-peeling-for-alignment.patch: Move to ...
- fix-PR-92351-When-peeling-for-alignment.patch: ... this
- AArch64-Fix-constraints-for-CPY-M.patch: New file
- Apply-maximum-nunits-for-BB-SLP.patch: New file
- Fix-EXTRACT_LAST_REDUCTION-segfault.patch: New file
- Fix-up-push_partial_def-little-endian-bitfield.patch: New file
- Fix-zero-masking-for-vcvtps2ph.patch: New file
- IRA-Handle-fully-tied-destinations.patch: New file
- SLP-VECT-Add-check-to-fix-96837.patch: New file
- aarch64-Fix-ash-lr-lshr-mode-3-expanders.patch: New file
- aarch64-Fix-bf16-and-matrix-g++-gfortran.patch: New file
- aarch64-Fix-mismatched-SVE-predicate-modes.patch: New file
- aarch64-fix-sve-acle-error.patch: New file
- adjust-vector-cost-and-move-EXTRACT_LAST_REDUCTION-costing.patch: New file
- bf16-and-matrix-characteristic.patch: New file
- fix-ICE-IPA-compare-VRP-types.patch: New file
- fix-ICE-in-affine-combination.patch: New file
- fix-ICE-in-pass-vect.patch: New file
- fix-ICE-in-vect_update_misalignment_for_peel.patch: New file
- fix-addlosymdi-ICE-in-pass-reload.patch: New file
- fix-an-ICE-in-vect_recog_mask_conversion_pattern.patch: New file
- fix-avx512vl-vcvttpd2dq-2-fail.patch: New file
- fix-issue499-add-nop-convert.patch: New file
- fix-issue604-ldist-dependency-fixup.patch: New file
- modulo-sched-Carefully-process-loop-counter-initiali.patch: New file
- re-PR-target-91124-gcc.target-i386-avx512vl-vpshldvd.patch: New file
- reduction-paths-with-unhandled-live-stmt.patch: New file
- redundant-loop-elimination.patch: New file
- sccvn-Improve-handling-of-load-masked-with-integer.patch: New file
- speed-up-DDG-analysis-and-fix-bootstrap-compare-debug.patch: New file
- store-merging-Consider-also-overlapping-stores-earlier.patch: New file
- tree-optimization-96920-another-ICE-when-vectorizing.patch: New file
- tree-optimization-97812-fix-range-query-in-VRP-asser.patch: New file
- vectorizable-comparison-Swap-operands-only-once.patch: New file
- x86-Fix-bf16-and-matrix.patch: New file
2020-12-30 09:54:10 +08:00

89 lines
3.3 KiB
Diff

This backport contains 2 patchs from gcc main stream tree.
The commit id of these patchs list as following in the order of time.
4bf29d15f2e01348a45a1f4e1a135962f123fdd6
0001-AArch64-PR79262-Adjust-vector-cost.patch
27071013521b015d17a2666448f27a6ff0c55aca
0001-Move-EXTRACT_LAST_REDUCTION-costing-to-vectorizable_.patch
diff -Nurp a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
--- a/gcc/config/aarch64/aarch64.c 2020-11-20 04:36:33.988000000 +0800
+++ b/gcc/config/aarch64/aarch64.c 2020-11-20 04:32:20.984000000 +0800
@@ -448,7 +448,7 @@ static const struct cpu_vector_cost gene
1, /* vec_int_stmt_cost */
1, /* vec_fp_stmt_cost */
2, /* vec_permute_cost */
- 1, /* vec_to_scalar_cost */
+ 2, /* vec_to_scalar_cost */
1, /* scalar_to_vec_cost */
1, /* vec_align_load_cost */
1, /* vec_unalign_load_cost */
diff -Nurp a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
--- a/gcc/tree-vect-loop.c 2020-11-20 04:36:34.016000000 +0800
+++ b/gcc/tree-vect-loop.c 2020-11-20 04:32:20.984000000 +0800
@@ -3926,8 +3926,11 @@ vect_model_reduction_cost (stmt_vec_info
code = gimple_assign_rhs_code (orig_stmt_info->stmt);
- if (reduction_type == EXTRACT_LAST_REDUCTION
- || reduction_type == FOLD_LEFT_REDUCTION)
+ if (reduction_type == EXTRACT_LAST_REDUCTION)
+ /* No extra instructions are needed in the prologue. The loop body
+ operations are costed in vectorizable_condition. */
+ inside_cost = 0;
+ else if (reduction_type == FOLD_LEFT_REDUCTION)
{
/* No extra instructions needed in the prologue. */
prologue_cost = 0;
diff -Nurp a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
--- a/gcc/tree-vect-stmts.c 2020-11-20 04:36:33.996000000 +0800
+++ b/gcc/tree-vect-stmts.c 2020-11-20 04:32:20.984000000 +0800
@@ -859,7 +859,8 @@ vect_model_simple_cost (stmt_vec_info st
enum vect_def_type *dt,
int ndts,
slp_tree node,
- stmt_vector_for_cost *cost_vec)
+ stmt_vector_for_cost *cost_vec,
+ vect_cost_for_stmt kind = vector_stmt)
{
int inside_cost = 0, prologue_cost = 0;
@@ -906,7 +907,7 @@ vect_model_simple_cost (stmt_vec_info st
}
/* Pass the inside-of-loop statements to the target-specific cost model. */
- inside_cost += record_stmt_cost (cost_vec, ncopies, vector_stmt,
+ inside_cost += record_stmt_cost (cost_vec, ncopies, kind,
stmt_info, 0, vect_body);
if (dump_enabled_p ())
@@ -9194,15 +9195,18 @@ vectorizable_condition (stmt_vec_info st
" EXTRACT_LAST_REDUCTION.\n");
LOOP_VINFO_CAN_FULLY_MASK_P (loop_vinfo) = false;
}
- if (expand_vec_cond_expr_p (vectype, comp_vectype,
- cond_code))
- {
- STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
- vect_model_simple_cost (stmt_info, ncopies, dts, ndts, slp_node,
- cost_vec);
- return true;
- }
- return false;
+
+ vect_cost_for_stmt kind = vector_stmt;
+ if (reduction_type == EXTRACT_LAST_REDUCTION)
+ /* Count one reduction-like operation per vector. */
+ kind = vec_to_scalar;
+ else if (!expand_vec_cond_expr_p (vectype, comp_vectype, cond_code))
+ return false;
+
+ STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type;
+ vect_model_simple_cost (stmt_info, ncopies, dts, ndts, slp_node,
+ cost_vec, kind);
+ return true;
}
/* Transform. */