From patchwork Tue Nov 11 14:21:08 2014 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Evgeny Stupachenko X-Patchwork-Id: 409467 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 8B05214012C for ; Wed, 12 Nov 2014 01:21:20 +1100 (AEDT) DomainKey-Signature: a=rsa-sha1; c=nofws; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender :mime-version:date:message-id:subject:from:to:content-type; q= dns; s=default; b=oISbgbkLH5bQ3eKsLgCylKoM7jB2DQDQHBmjDFm6CgUmYj XUWk9HB/Q5eXQBhs8gSPp+GEbIKdlwS8eDH3TE6gQcvstWZqfvBRgSrHh9Qm4r86 ZB8XiPLa86W6Bq1PtSSy7M0int6NApclvSqqwNkg5ZaYF0+hPFUHvsjLvDYRc= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender :mime-version:date:message-id:subject:from:to:content-type; s= default; bh=9gSsc0gw7PIT0BZ//L0e/XvrNGw=; b=ttMoTBhskZOk+Y8MMYVo BT2klgEv0ECic+QLOVN9SHwWUZB7bVo3bpgN70mkg0E04ClHwCs47capkpT918vt 7KmgG/ohUhi4tOb2pPI/Hqcijenh60Inhhpd0wPvV8CMSyaWIiwC2FGmCviqj5sP nVod1/aikDqqiGrcqTTFDAY= Received: (qmail 23208 invoked by alias); 11 Nov 2014 14:21:13 -0000 Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org Received: (qmail 23197 invoked by uid 89); 11 Nov 2014 14:21:12 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-2.5 required=5.0 tests=AWL, BAYES_00, FREEMAIL_FROM, RCVD_IN_DNSWL_LOW, SPF_PASS autolearn=ham version=3.3.2 X-HELO: mail-ie0-f173.google.com Received: from mail-ie0-f173.google.com (HELO mail-ie0-f173.google.com) (209.85.223.173) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with (AES128-SHA encrypted) ESMTPS; Tue, 11 Nov 2014 14:21:11 +0000 Received: by mail-ie0-f173.google.com with SMTP id tr6so11480769ieb.32 for ; Tue, 11 Nov 2014 06:21:09 -0800 (PST) MIME-Version: 1.0 X-Received: by 10.107.3.163 with SMTP id e35mr42144819ioi.45.1415715668860; Tue, 11 Nov 2014 06:21:08 -0800 (PST) Received: by 10.107.11.220 with HTTP; Tue, 11 Nov 2014 06:21:08 -0800 (PST) Date: Tue, 11 Nov 2014 17:21:08 +0300 Message-ID: Subject: [PATCH] Extend shift permutations on power of 2 cases From: Evgeny Stupachenko To: GCC Patches , Richard Biener , Richard Henderson , Uros Bizjak X-IsSubscribed: yes Hi, The patch extends shift permutations technique on power of 2 cases (previously even/odd transformations was used unconditionally). Basically the patch just add loop for load group of length 2, like it is done in "vect_permute_load_chain" function. For Silvermont it reduces insn sequence for load group of length 4 from 31 to 20 insns. Performance for the test in the patch improved by ~20%. Bootstrap passed. Make check in progress. Is it ok? 2014-11-11 Evgeny Stupachenko gcc/testsuite * gcc.target/i386/pr52252-atom-1.c: New. gcc/ * tree-vect-data-refs.c (vect_shift_permute_load_chain): Extend shift permutations on power of 2 cases. if (length == 3 && LOOP_VINFO_VECT_FACTOR (loop_vinfo) > 2) diff --git a/gcc/testsuite/gcc.target/i386/pr52252-atom-1.c b/gcc/testsuite/gcc.target/i386/pr52252-atom-1.c new file mode 100644 index 0000000..1fbd258 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr52252-atom-1.c @@ -0,0 +1,22 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target ssse3 } */ +/* { dg-options "-O2 -ftree-vectorize -mssse3 -mtune=slm" } */ +#define byte unsigned char + +void +pair_mul_sum(byte *in, byte *out, int size) +{ + int j; + for(j = 0; j < size; j++) + { + byte a = in[0]; + byte b = in[1]; + byte c = in[2]; + byte d = in[3]; + out[0] = (byte)(a * b) + (byte)(b * c) + (byte)(c * d) + (byte)(d * a); + in += 4; + out += 1; + } +} + +/* { dg-final { scan-assembler "palignr" } } */ diff --git a/gcc/tree-vect-data-refs.c b/gcc/tree-vect-data-refs.c index 0bc0356..d2e0e93 100644 --- a/gcc/tree-vect-data-refs.c +++ b/gcc/tree-vect-data-refs.c @@ -5379,8 +5379,9 @@ vect_shift_permute_load_chain (vec dr_chain, memcpy (result_chain->address (), dr_chain.address (), length * sizeof (tree)); - if (length == 2 && LOOP_VINFO_VECT_FACTOR (loop_vinfo) > 4) + if (exact_log2 (length) != -1 && LOOP_VINFO_VECT_FACTOR (loop_vinfo) > 4) { + unsigned int j, log_length = exact_log2 (length); for (i = 0; i < nelt / 2; ++i) sel[i] = i * 2; for (i = 0; i < nelt / 2; ++i) @@ -5441,37 +5442,44 @@ vect_shift_permute_load_chain (vec dr_chain, select_mask = vect_gen_perm_mask (vectype, sel); gcc_assert (select_mask != NULL); - first_vect = dr_chain[0]; - second_vect = dr_chain[1]; - - data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2"); - perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, - first_vect, first_vect, - perm2_mask1); - vect_finish_stmt_generation (stmt, perm_stmt, gsi); - vect[0] = data_ref; + for (i = 0; i < log_length; i++) + { + for (j = 0; j < length; j += 2) + { + first_vect = dr_chain[j]; + second_vect = dr_chain[j + 1]; - data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2"); - perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, - second_vect, second_vect, - perm2_mask2); - vect_finish_stmt_generation (stmt, perm_stmt, gsi); - vect[1] = data_ref; + data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2"); + perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, + first_vect, first_vect, + perm2_mask1); + vect_finish_stmt_generation (stmt, perm_stmt, gsi); + vect[0] = data_ref; - data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift"); - perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, - vect[0], vect[1], - shift1_mask); - vect_finish_stmt_generation (stmt, perm_stmt, gsi); - (*result_chain)[1] = data_ref; + data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2"); + perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, + second_vect, second_vect, + perm2_mask2); + vect_finish_stmt_generation (stmt, perm_stmt, gsi); + vect[1] = data_ref; - data_ref = make_temp_ssa_name (vectype, NULL, "vect_select"); - perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, - vect[0], vect[1], - select_mask); - vect_finish_stmt_generation (stmt, perm_stmt, gsi); - (*result_chain)[0] = data_ref; + data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift"); + perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, + vect[0], vect[1], + shift1_mask); + vect_finish_stmt_generation (stmt, perm_stmt, gsi); + (*result_chain)[j/2 + length/2] = data_ref; + data_ref = make_temp_ssa_name (vectype, NULL, "vect_select"); + perm_stmt = gimple_build_assign_with_ops (VEC_PERM_EXPR, data_ref, + vect[0], vect[1], + select_mask); + vect_finish_stmt_generation (stmt, perm_stmt, gsi); + (*result_chain)[j/2] = data_ref; + } + memcpy (dr_chain.address (), result_chain->address (), + length * sizeof (tree)); + } return true; }