gcc/aarch64-Fix-mismatched-SVE-predicate-modes.patch

35 lines
1.4 KiB
Diff
Raw Normal View History

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
This backport contains 1 patch from gcc main stream tree.
The commit id of these patchs list as following in the order of time.
0001-aarch64-Fix-mismatched-SVE-predicate-modes.patch
26bebf576ddcdcfb596f07e8c2896f17c48516e7
diff -urpN a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
--- a/gcc/config/aarch64/aarch64.c 2020-12-14 00:57:20.128000000 -0500
+++ b/gcc/config/aarch64/aarch64.c 2020-12-14 01:00:15.080000000 -0500
@@ -4328,6 +4328,7 @@ aarch64_expand_sve_const_pred_eor (rtx t
/* EOR the result with an ELT_SIZE PTRUE. */
rtx mask = aarch64_ptrue_all (elt_size);
mask = force_reg (VNx16BImode, mask);
+ inv = gen_lowpart (VNx16BImode, inv);
target = aarch64_target_reg (target, VNx16BImode);
emit_insn (gen_aarch64_pred_z (XOR, VNx16BImode, target, mask, inv, mask));
return target;
diff -urpN a/gcc/testsuite/gcc.dg/vect/pr94606.c b/gcc/testsuite/gcc.dg/vect/pr94606.c
--- a/gcc/testsuite/gcc.dg/vect/pr94606.c 1969-12-31 19:00:00.000000000 -0500
+++ b/gcc/testsuite/gcc.dg/vect/pr94606.c 2020-12-14 01:00:15.080000000 -0500
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-march=armv8.2-a+sve -msve-vector-bits=256" { target aarch64*-*-* } } */
+
+const short mask[] = { 0, 0, 0, 0, 0, 0, 0, 0,
+ 0, 0, 0, 1, 1, 1, 1, 1 };
+
+int
+foo (short *restrict x, short *restrict y)
+{
+ for (int i = 0; i < 16; ++i)
+ if (mask[i])
+ x[i] += y[i];
+}