[AArch64/GCC,17/N] Optimize prologue when there is no frame pointe

Message ID	53D0FF7E.1070309@arm.com
State	New
Headers	show Return-Path: <gcc-patches-return-373196-incoming=patchwork.ozlabs.org@gcc.gnu.org> DomainKey-Signature: a=rsa-sha1; c=nofws; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender :message-id:date:from:mime-version:to:subject:content-type; q= dns; s=default; b=GqLphyy03IzNlQif591RKBlQhnvBi9rVy5N1H/kQKd2vGN eQG4Ea3pmMgMUcPE2K1zBGG0/st2qMbP0PyK/Hn+9Nb3U9v7jBmvKxPa9RW0iiKA MJVbhbQ0It4Bont3tExb4eHZmlpfZtiqUd1XKx1f8wC61P0xlUmlWIvquNlwI= Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk Sender: gcc-patches-owner@gcc.gnu.org Message-ID: <53D0FF7E.1070309@arm.com> Date: Thu, 24 Jul 2014 13:43:42 +0100 From: Jiong Wang <jiong.wang@arm.com> User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.1.1 MIME-Version: 1.0 To: "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org> Subject: [AArch64/GCC][17/N] Optimize prologue when there is no frame pointe Content-Type: multipart/mixed; boundary="------------000909010801080702050600"

From e3ab087747c2f4ddeef0482983b2ebc3bbdc131f Mon Sep 17 00:00:00 2001 From: Jiong Wang <jiong.wang@arm.com> Date: Tue, 17 Jun 2014 22:24:44 +0100 Subject: [PATCH 17/19] [AArch64/GCC][18/20] Optimize prologue when there is no frame pointer Under new pro/epi code, we could also utilize our store write-back to optimize stack adjustment when there is no frame pointer. * if there is candidate reg pair and adjustment amount is less than 512 then we could use aarch64's paired store write-back. * if there is only a single candidate reg and adjustment amount is less than 256, we could use aarch64's single store write-back. * otherwise use explictly subtraction to finish stack adjustment. Improved testcases: gcc.target/aarch64/test_frame_1.c gcc.target/aarch64/test_frame_10.c gcc.target/aarch64/test_frame_2.c gcc.target/aarch64/test_frame_4.c gcc.target/aarch64/test_frame_6.c gcc.target/aarch64/test_frame_7.c gcc.target/aarch64/test_frame_8.c gcc.target/aarch64/test_fp_attribute_1.c 2014-06-16 Jiong Wang <jiong.wang@arm.com> Marcus Shawcroft <marcus.shawcroft@arm.com> gcc/ * config/aarch64/aarch64.c (aarch64_pushwb_single_reg): New function. (aarch64_expand_prologue): Optimize prologue when !frame_pointer_needed. gcc/testsuite/ * gcc.target/aarch64/test_frame_1.c: Match optimized instruction sequences. * gcc.target/aarch64/test_frame_10.c: Likewise. * gcc.target/aarch64/test_frame_2.c: Likewise. * gcc.target/aarch64/test_frame_4.c: Likewise. * gcc.target/aarch64/test_frame_6.c: Likewise. * gcc.target/aarch64/test_frame_7.c: Likewise. * gcc.target/aarch64/test_frame_8.c: Likewise. * gcc.target/aarch64/test_fp_attribute_1.c: Likewise. --- gcc/config/aarch64/aarch64.c | 58 +++++++++++++++----- .../gcc.target/aarch64/test_fp_attribute_1.c | 2 +- gcc/testsuite/gcc.target/aarch64/test_frame_1.c | 5 +- gcc/testsuite/gcc.target/aarch64/test_frame_10.c | 5 +- gcc/testsuite/gcc.target/aarch64/test_frame_2.c | 6 +- gcc/testsuite/gcc.target/aarch64/test_frame_4.c | 5 +- gcc/testsuite/gcc.target/aarch64/test_frame_6.c | 5 +- gcc/testsuite/gcc.target/aarch64/test_frame_7.c | 5 +- gcc/testsuite/gcc.target/aarch64/test_frame_8.c | 5 +- 9 files changed, 74 insertions(+), 22 deletions(-) diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c index 26d5fba..365fdd4 100644 --- a/gcc/config/aarch64/aarch64.c +++ b/gcc/config/aarch64/aarch64.c @@ -1927,6 +1927,22 @@ aarch64_next_callee_save (unsigned regno, unsigned limit) return regno; } +static void +aarch64_pushwb_single_reg (enum machine_mode mode, unsigned regno, + HOST_WIDE_INT adjustment) + { + rtx base_rtx = stack_pointer_rtx; + rtx insn, reg, mem; + + reg = gen_rtx_REG (mode, regno); + mem = gen_rtx_PRE_MODIFY (Pmode, base_rtx, + plus_constant (Pmode, base_rtx, -adjustment)); + mem = gen_rtx_MEM (mode, mem); + + insn = emit_move_insn (mem, reg); + RTX_FRAME_RELATED_P (insn) = 1; +} + static rtx aarch64_gen_storewb_pair (enum machine_mode mode, rtx base, rtx reg, rtx reg2, HOST_WIDE_INT adjustment) @@ -2276,11 +2292,10 @@ aarch64_expand_prologue (void) { bool skip_wb = false; - /* Save the frame pointer and lr if the frame pointer is needed - first. Make the frame pointer point to the location of the - old frame pointer on the stack. */ if (frame_pointer_needed) { + skip_wb = true; + if (fp_offset) { insn = emit_insn (gen_add2_insn (stack_pointer_rtx, @@ -2288,12 +2303,11 @@ aarch64_expand_prologue (void) RTX_FRAME_RELATED_P (insn) = 1; aarch64_set_frame_expr (gen_rtx_SET (Pmode, stack_pointer_rtx, - gen_rtx_MINUS (Pmode, - stack_pointer_rtx, + gen_rtx_MINUS (Pmode, stack_pointer_rtx, GEN_INT (offset)))); aarch64_save_callee_saves (DImode, fp_offset, R29_REGNUM, - R30_REGNUM, skip_wb); + R30_REGNUM, false); } else aarch64_pushwb_pair_reg (DImode, R29_REGNUM, R30_REGNUM, offset); @@ -2311,20 +2325,36 @@ aarch64_expand_prologue (void) RTX_FRAME_RELATED_P (insn) = 1; insn = emit_insn (gen_stack_tie (stack_pointer_rtx, hard_frame_pointer_rtx)); - - aarch64_save_callee_saves (DImode, fp_offset, R0_REGNUM, R28_REGNUM, - skip_wb); } else { - insn = emit_insn (gen_add2_insn (stack_pointer_rtx, - GEN_INT (-offset))); - RTX_FRAME_RELATED_P (insn) = 1; + unsigned reg1 = cfun->machine->frame.wb_candidate1; + unsigned reg2 = cfun->machine->frame.wb_candidate2; - aarch64_save_callee_saves (DImode, fp_offset, R0_REGNUM, R30_REGNUM, - skip_wb); + if (fp_offset + || reg1 == FIRST_PSEUDO_REGISTER + || (reg2 == FIRST_PSEUDO_REGISTER + && offset >= 256)) + { + insn = emit_insn (gen_add2_insn (stack_pointer_rtx, + GEN_INT (-offset))); + RTX_FRAME_RELATED_P (insn) = 1; + } + else + { + enum machine_mode mode1 = (reg1 <= R30_REGNUM) ? DImode : DFmode; + + skip_wb = true; + + if (reg2 == FIRST_PSEUDO_REGISTER) + aarch64_pushwb_single_reg (mode1, reg1, offset); + else + aarch64_pushwb_pair_reg (mode1, reg1, reg2, offset); + } } + aarch64_save_callee_saves (DImode, fp_offset, R0_REGNUM, R30_REGNUM, + skip_wb); aarch64_save_callee_saves (DFmode, fp_offset, V0_REGNUM, V31_REGNUM, skip_wb); } diff --git a/gcc/testsuite/gcc.target/aarch64/test_fp_attribute_1.c b/gcc/testsuite/gcc.target/aarch64/test_fp_attribute_1.c index 7538250..960174a 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_fp_attribute_1.c +++ b/gcc/testsuite/gcc.target/aarch64/test_fp_attribute_1.c @@ -21,6 +21,6 @@ non_leaf_2 (void) leaf (); } -/* { dg-final { scan-assembler-times "str\tx30, \\\[sp\\\]" 2 } } */ +/* { dg-final { scan-assembler-times "str\tx30, \\\[sp, -\[0-9\]+\\\]!" 2 } } */ /* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_1.c b/gcc/testsuite/gcc.target/aarch64/test_frame_1.c index feea7a2..e9d04aa 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_1.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_1.c @@ -6,9 +6,12 @@ * optimized code should use "str !" for stack adjustment. */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern (test1, 200, ) t_frame_run (test1) + +/* { dg-final { scan-assembler-times "str\tx30, \\\[sp, -\[0-9\]+\\\]!" 2 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_10.c b/gcc/testsuite/gcc.target/aarch64/test_frame_10.c index 2892c5f..b646a71 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_10.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_10.c @@ -8,9 +8,12 @@ the first subtractions could be optimized into "stp !". */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern_outgoing (test10, 480, "x19", 24, a[8], a[9], a[10]) t_frame_run (test10) + +/* { dg-final { scan-assembler-times "stp\tx19, x30, \\\[sp, -\[0-9\]+\\\]!" 1 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_2.c b/gcc/testsuite/gcc.target/aarch64/test_frame_2.c index aa15dae..b972664 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_2.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_2.c @@ -6,9 +6,13 @@ * optimized code should use "stp !" for stack adjustment. */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern (test2, 200, "x19") t_frame_run (test2) + + +/* { dg-final { scan-assembler-times "stp\tx19, x30, \\\[sp, -\[0-9\]+\\\]!" 1 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_4.c b/gcc/testsuite/gcc.target/aarch64/test_frame_4.c index c45e740..5a9a919 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_4.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_4.c @@ -6,9 +6,12 @@ * we can use "stp !" to optimize stack adjustment. */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern (test4, 400, "x19") t_frame_run (test4) + +/* { dg-final { scan-assembler-times "stp\tx19, x30, \\\[sp, -\[0-9\]+\\\]!" 1 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_6.c b/gcc/testsuite/gcc.target/aarch64/test_frame_6.c index 54f646b..6056f57 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_6.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_6.c @@ -7,9 +7,12 @@ the second subtraction should use "str !". */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern (test6, 700, ) t_frame_run (test6) + +/* { dg-final { scan-assembler-times "str\tx30, \\\[sp, -\[0-9\]+\\\]!" 2 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_7.c b/gcc/testsuite/gcc.target/aarch64/test_frame_7.c index aa97bc0..991860c 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_7.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_7.c @@ -7,9 +7,12 @@ the second subtraction should use "stp !". */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern (test7, 700, "x19") t_frame_run (test7) + +/* { dg-final { scan-assembler-times "stp\tx19, x30, \\\[sp, -\[0-9\]+\\\]!" 1 } } */ +/* { dg-final { cleanup-saved-temps } } */ diff --git a/gcc/testsuite/gcc.target/aarch64/test_frame_8.c b/gcc/testsuite/gcc.target/aarch64/test_frame_8.c index f75f080..4a4d93b 100644 --- a/gcc/testsuite/gcc.target/aarch64/test_frame_8.c +++ b/gcc/testsuite/gcc.target/aarch64/test_frame_8.c @@ -5,9 +5,12 @@ * number of callee-saved reg == 1. */ /* { dg-do run } */ -/* { dg-options "-O2 -fomit-frame-pointer" } */ +/* { dg-options "-O2 -fomit-frame-pointer --save-temps" } */ #include "test_frame_common.h" t_frame_pattern_outgoing (test8, 700, , 8, a[8]) t_frame_run (test8) + +/* { dg-final { scan-assembler-times "str\tx30, \\\[sp, -\[0-9\]+\\\]!" 3 } } */ +/* { dg-final { cleanup-saved-temps } } */ -- 1.7.9.5

[AArch64/GCC,17/N] Optimize prologue when there is no frame pointe

Commit Message

Comments

Patch