From 546e705f38aa5c125cc29469d3c8f1b789cf788a Mon Sep 17 00:00:00 2001 From: ienkovich Date: Mon, 14 Sep 2015 09:36:29 +0000 Subject: [PATCH] * config/i386/haswell.md: New file describing Haswell pipeline. * config/i386/i386.c (processor_alias_table): Use CPU_HASWELL for haswell-like processors. (ix86_reassociation_width): Increase reassociation width for 64-bit Haswell processor family. * config/i386/i386.md: Introduce haswell cpu and include new md file. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@227740 138bc75d-0d04-0410-961f-82ee72b054a4 --- gcc/ChangeLog | 9 + gcc/config/i386/haswell.md | 618 +++++++++++++++++++++++++++++++++++++++++++++ gcc/config/i386/i386.c | 10 +- gcc/config/i386/i386.md | 5 +- 4 files changed, 635 insertions(+), 7 deletions(-) create mode 100644 gcc/config/i386/haswell.md diff --git a/gcc/ChangeLog b/gcc/ChangeLog index e347f03..7842cfe 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,12 @@ +2015-09-14 Yuri Rumyantsev + + * config/i386/haswell.md: New file describing Haswell pipeline. + * config/i386/i386.c (processor_alias_table): Use CPU_HASWELL for + haswell-like processors. + (ix86_reassociation_width): Increase reassociation width for 64-bit + Haswell processor family. + * config/i386/i386.md: Introduce haswell cpu and include new md file. + 2015-09-14 Richard Biener * doc/match-and-simplify.texi: Fixup some formatting issues diff --git a/gcc/config/i386/haswell.md b/gcc/config/i386/haswell.md new file mode 100644 index 0000000..2bb0ac9 --- /dev/null +++ b/gcc/config/i386/haswell.md @@ -0,0 +1,618 @@ +;; Scheduling for Haswell and derived processors. +;; Copyright (C) 2004-2015 Free Software Foundation, Inc. +;; +;; This file is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify +;; it under the terms of the GNU General Public License as published by +;; the Free Software Foundation; either version 3, or (at your option) +;; any later version. +;; +;; GCC is distributed in the hope that it will be useful, +;; but WITHOUT ANY WARRANTY; without even the implied warranty of +;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;; GNU General Public License for more details. +;; +;; You should have received a copy of the GNU General Public License +;; along with GCC; see the file COPYING3. If not see +;; . */ + +;; The scheduling description in this file is based on core2.md. +;; The major difference from the CORE2 pipeline is that HASWELL has +;; two MU for load and one MU for store. +(define_automaton "haswell_decoder,haswell_core,haswell_idiv,haswell_fdiv,haswell_ssediv,haswell_load,haswell_store") + +;; The CPU domain, used for HASWELL bypass latencies +(define_attr "hsw_domain" "int,float,simd" + (cond [(eq_attr "type" "fmov,fop,fsgn,fmul,fdiv,fpspc,fcmov,fcmp,fxch,fistp,fisttp,frndint") + (const_string "float") + (eq_attr "type" "sselog,sselog1,sseiadd,sseiadd1,sseishft,sseishft1,sseimul, + sse,ssemov,sseadd,sseadd1,ssemul,ssecmp,ssecomi,ssecvt, + ssecvt1,sseicvt,ssediv,sseins,ssemuladd,sse4arg") + (cond [(eq_attr "mode" "V4DF,V8SF,V2DF,V4SF,SF,DF") + (const_string "float") + (eq_attr "mode" "SI") + (const_string "int")] + (const_string "simd")) + (eq_attr "type" "mmx,mmxmov,mmxadd,mmxmul,mmxcmp,mmxcvt,mmxshft") + (const_string "simd")] + (const_string "int"))) + +(define_cpu_unit "hsw_decoder0" "haswell_decoder") +(define_cpu_unit "hsw_decoder1" "haswell_decoder") +(define_cpu_unit "hsw_decoder2" "haswell_decoder") +(define_cpu_unit "hsw_decoder3" "haswell_decoder") + +;; We first wish to find an instruction for hsw_decoder0, so exclude +;; other hsw_decoders from being reserved until hsw_decoder0 is +;; reserved. +(presence_set "hsw_decoder1" "hsw_decoder0") +(presence_set "hsw_decoder2" "hsw_decoder0") +(presence_set "hsw_decoder3" "hsw_decoder0") + +;; Most instructions can be decoded on any of the three decoders. +(define_reservation "hsw_decodern" "(hsw_decoder0|hsw_decoder1|hsw_decoder2|hsw_decoder3)") + +;; The out-of-order core has eight pipelines. These are similar to the +;; Pentium Pro's five pipelines. Port 2,3 are responsible for memory loads, +;; port 7 for store address calculations, port 4 for memory stores, and +;; ports 0, 1, 5 and 6 for everything else. + +(define_cpu_unit "hsw_p0,hsw_p1,hsw_p5,hsw_p6" "haswell_core") +(define_cpu_unit "hsw_p2,hsw_p3" "haswell_load") +(define_cpu_unit "hsw_p4,hsw_p7" "haswell_store") +(define_cpu_unit "hsw_idiv" "haswell_idiv") +(define_cpu_unit "hsw_fdiv" "haswell_fdiv") +(define_cpu_unit "hsw_ssediv" "haswell_ssediv") + +(define_reservation "hsw_p0156" "hsw_p0|hsw_p1|hsw_p5|hsw_p6") +(define_reservation "hsw_p0p1p5p6" "hsw_p0+hsw_p1+hsw_p5+hsw_p6") +(define_reservation "hsw_p23" "hsw_p2|hsw_p3") +(define_reservation "hsw_p4p7" "hsw_p4+hsw_p7") +(define_reservation "hsw_p237" "hsw_p2|hsw_p3|hsw_p7") +(define_reservation "hsw_p015" "hsw_p0|hsw_p1|hsw_p5") +(define_reservation "hsw_p01" "hsw_p0|hsw_p1") + +(define_insn_reservation "hsw_complex_insn" 6 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "other,multi,str")) + "hsw_decoder0") + +(define_insn_reservation "hsw_call" 1 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "call,callv")) + "hsw_decoder0") + +;; imov with memory operands does not use the integer units. +;; imovx always decodes to one uop, and also doesn't use the integer +;; units if it has memory operands. +(define_insn_reservation "hsw_imov" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "imov,imovx"))) + "hsw_decodern,hsw_p0156") + +(define_insn_reservation "hsw_imov_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "imov,imovx"))) + "hsw_decodern,hsw_p23") + +(define_insn_reservation "hsw_imov_store" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (eq_attr "type" "imov"))) + "hsw_decodern,hsw_p4+(hsw_p2|hsw_p3|hsw_p7)") + +(define_insn_reservation "hsw_icmov" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "icmov"))) + "hsw_decodern,hsw_p0156,hsw_p0156") + +(define_insn_reservation "hsw_icmov_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "icmov"))) + "hsw_decodern,hsw_p23+hsw_p0156,hsw_p0156") + +(define_insn_reservation "hsw_push_reg" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (eq_attr "type" "push"))) + "hsw_decodern,hsw_p4+hsw_p237") + +(define_insn_reservation "hsw_push_mem" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "both") + (eq_attr "type" "push"))) + "hsw_decodern,hsw_p4+hsw_p237,hsw_p237") + +;; Consider lea latency as having 2 components. +(define_insn_reservation "hsw_lea" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "lea"))) + "hsw_decodern,hsw_p1|hsw_p5") + +(define_insn_reservation "hsw_shift_rotate" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "ishift,ishift1,rotate,rotate1"))) + "hsw_decodern,hsw_p0|hsw_p6") + +(define_insn_reservation "hsw_shift_rotate_mem" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (eq_attr "type" "ishift,ishift1,rotate,rotate1"))) + "hsw_decodern,(hsw_p0|hsw_p6)+hsw_p237+hsw_p4") + +(define_insn_reservation "hsw_branch" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "ibr"))) + "hsw_decodern,hsw_p6") + +(define_insn_reservation "hsw_indirect_branch" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (eq_attr "type" "ibr"))) + "hsw_decoder0,hsw_p23+hsw_p6") + +(define_insn_reservation "hsw_leave" 4 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "leave")) + "hsw_decoder0,hsw_p23+hsw_p0156,hsw_p0156") + +;; imul and imulx with two/three operands only execute on port 1. +(define_insn_reservation "hsw_imul" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "imul"))) + "hsw_decodern,hsw_p1") + +(define_insn_reservation "hsw_imul_mem" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (eq_attr "type" "imul"))) + "hsw_decodern,hsw_p23+hsw_p1") + +(define_insn_reservation "hsw_imulx" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "imulx"))) + "hsw_decodern,hsw_p0156,hsw_p0156") + +(define_insn_reservation "hsw_imulx_mem" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (eq_attr "type" "imulx"))) + "hsw_decodern,hsw_p23+hsw_p0156,(hsw_p0|hsw_p6|hsw_p6)") + + +;; div and idiv are very similar, so we model them the same. +;; Use the same latency for all QI,HI and SI modes. +(define_insn_reservation "hsw_idiv" 23 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "idiv"))) + "hsw_decoder0,(hsw_p0p1p5p6+hsw_idiv)*9") + +(define_insn_reservation "hsw_idiv_load" 23 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "idiv"))) + "hsw_decoder0,hsw_p23+hsw_p0+hsw_idiv,(hsw_p0p1p5p6+hsw_idiv)*9") + +;; x87 floating point operations. + +(define_insn_reservation "hsw_fxch" 0 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "fxch")) + "hsw_decodern") + +(define_insn_reservation "hsw_fop" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none,unknown") + (eq_attr "type" "fop"))) + "hsw_decodern,hsw_p1") + +(define_insn_reservation "hsw_fop_load" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "fop"))) + "hsw_decodern,hsw_p23+hsw_p1,hsw_p1") + +(define_insn_reservation "hsw_fop_store" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (eq_attr "type" "fop"))) + "hsw_decodern,hsw_p0,hsw_p0,hsw_p0+hsw_p4+hsw_p3") + +(define_insn_reservation "hsw_fop_both" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "both") + (eq_attr "type" "fop"))) + "hsw_decodern,hsw_p2+hsw_p0,hsw_p0+hsw_p4+hsw_p3") + +(define_insn_reservation "hsw_fsgn" 1 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "fsgn")) + "hsw_decodern,hsw_p0") + +(define_insn_reservation "hsw_fistp" 7 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "fistp")) + "hsw_decoder0,hsw_p1+hsw_p4+hsw_p23") + +(define_insn_reservation "hsw_fcmov" 2 + (and (eq_attr "cpu" "haswell") + (eq_attr "type" "fcmov")) + "hsw_decoder0,hsw_p0+hsw_p5,hsw_p0") + +(define_insn_reservation "hsw_fcmp" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "fcmp"))) + "hsw_decodern,hsw_p1") + +(define_insn_reservation "hsw_fcmp_load" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "fcmp"))) + "hsw_decodern,hsw_p23+hsw_p1") + +(define_insn_reservation "hsw_fmov" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "fmov"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_fmov_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "mode" "!XF") + (eq_attr "type" "fmov")))) + "hsw_decodern,hsw_p23") + +(define_insn_reservation "hsw_fmov_XF_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "mode" "XF") + (eq_attr "type" "fmov")))) + "hsw_decodern,(hsw_p23+hsw_p0)*2") + +(define_insn_reservation "hsw_fmov_store" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (and (eq_attr "mode" "!XF") + (eq_attr "type" "fmov")))) + "hsw_decodern,hsw_p4p7") + +(define_insn_reservation "hsw_fmov_XF_store" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (and (eq_attr "mode" "XF") + (eq_attr "type" "fmov")))) + "hsw_decodern,hsw_p4p7,hsw_p4p7") + +(define_insn_reservation "hsw_fmul" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "fmul"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_fmul_load" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "fmul"))) + "hsw_decodern,hsw_p23+hsw_p01") + +;; fdiv latencies depend on the mode of the operands. XFmode gives +;; a latency of 38 cycles, DFmode gives 32, and SFmode gives latency 18. +;; Division by a power of 2 takes only 9 cycles, but we cannot model +;; that. Throughput is equal to latency - 1, which we model using the +;; hsw_div automaton. +(define_insn_reservation "hsw_fdiv_SF" 18 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "SF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p0+hsw_fdiv,hsw_fdiv*16") + +(define_insn_reservation "hsw_fdiv_SF_load" 19 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "mode" "SF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p23+hsw_p0+hsw_fdiv,hsw_fdiv*16") + +(define_insn_reservation "hsw_fdiv_DF" 32 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "DF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p0+hsw_fdiv,hsw_fdiv*30") + +(define_insn_reservation "hsw_fdiv_DF_load" 33 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "mode" "DF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p23+hsw_p0+hsw_fdiv,hsw_fdiv*30") + +(define_insn_reservation "hsw_fdiv_XF" 38 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "XF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p0+hsw_fdiv,hsw_fdiv*36") + +(define_insn_reservation "hsw_fdiv_XF_load" 39 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "mode" "XF") + (eq_attr "type" "fdiv,fpspc")))) + "hsw_decodern,hsw_p2+hsw_p0+hsw_fdiv,hsw_fdiv*36") + +;; MMX instructions. + +(define_insn_reservation "hsw_mmx_add" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "mmxadd,sseiadd"))) + "hsw_decodern,hsw_p1|hsw_p5") + +(define_insn_reservation "hsw_mmx_add_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "mmxadd,sseiadd"))) + "hsw_decodern,hsw_p23+(hsw_p1|hsw_p5)") + +(define_insn_reservation "hsw_mmx_shft" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "mmxshft"))) + "hsw_decodern,hsw_p0") + +(define_insn_reservation "hsw_mmx_shft_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "mmxshft"))) + "hsw_decodern,hsw_p23+hsw_p0") + +(define_insn_reservation "hsw_mmx_sse_shft" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "type" "sseishft") + (eq_attr "length_immediate" "!0")))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_mmx_sse_shft_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "type" "sseishft") + (eq_attr "length_immediate" "!0")))) + "hsw_decodern,hsw_p01+hsw_p23") + +(define_insn_reservation "hsw_mmx_sse_shft1" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "type" "sseishft") + (eq_attr "length_immediate" "0")))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_mmx_sse_shft1_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (and (eq_attr "type" "sseishft") + (eq_attr "length_immediate" "0")))) + "hsw_decodern,hsw_p01+hsw_p23") + +(define_insn_reservation "hsw_mmx_mul" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "mmxmul,sseimul"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_mmx_mul_load" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "mmxmul,sseimul"))) + "hsw_decodern,hsw_p23+hsw_p01") + +(define_insn_reservation "hsw_sse_mmxcvt" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "mode" "DI") + (eq_attr "type" "mmxcvt"))) + "hsw_decodern,hsw_p1") + +;; (define_insn_reservation "hsw_sse_mmxshft" 2 +;; (and (eq_attr "cpu" "haswell") +;; (and (eq_attr "mode" "TI") +;; (eq_attr "type" "mmxshft"))) +;; "hsw_decodern,hsw_p01") + +;; The sfence instruction. +(define_insn_reservation "hsw_sse_sfence" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "unknown") + (eq_attr "type" "sse"))) + "hsw_decoder0,hsw_p23+hsw_p4") + +(define_insn_reservation "hsw_sse_SFDF" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "mode" "SF,DF") + (eq_attr "type" "sse"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_sse_V4SF" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "mode" "V4SF") + (eq_attr "type" "sse"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_sse_V8SF" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "mode" "V8SF,V4DF") + (eq_attr "type" "sse"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_sse_addcmp" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "sseadd1,ssecmp,ssecomi"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_sse_addcmp_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "sseadd1,ssecmp,ssecomi"))) + "hsw_decodern,hsw_p23+hsw_p01") + +(define_insn_reservation "hsw_sse_logic" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "sselog,sselog1"))) + "hsw_decodern,hsw_p015") + +(define_insn_reservation "hsw_sse_logic_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "sselog,sselog1"))) + "hsw_decodern,hsw_p015+hsw_p23") + +(define_insn_reservation "hsw_sse_add" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "sseadd"))) + "hsw_decodern,hsw_p1|hsw_p5") + +(define_insn_reservation "hsw_sse_add_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "sseadd"))) + "hsw_decodern,(hsw_p1|hsw_p5)+hsw_p23") + +(define_insn_reservation "hsw_sse_mul" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "ssemul"))) + "hsw_decodern,hsw_p0") + +(define_insn_reservation "hsw_sse_mul_load" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "ssemul"))) + "hsw_decodern,hsw_p0+hsw_p23") +;; Use skylake pipeline. +(define_insn_reservation "hsw_sse_muladd" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "ssemuladd"))) + "hsw_decodern,hsw_p01") + +(define_insn_reservation "hsw_sse_muladd_load" 5 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "ssemuladd"))) + "hsw_decodern,hsw_p01+hsw_p23") + +(define_insn_reservation "hsw_sse_div_SF" 18 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "SF,V4SF,V8SF") + (eq_attr "type" "ssediv")))) + "hsw_decodern,hsw_p0,hsw_ssediv*14") + +(define_insn_reservation "hsw_sse_div_SF_load" 18 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "SF,V4SF,V8SF") + (eq_attr "type" "ssediv")))) + "hsw_decodern,(hsw_p23+hsw_p0),hsw_ssediv*14") + +(define_insn_reservation "hsw_sse_div_DF" 28 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "DF,V2DF,V4DF") + (eq_attr "type" "ssediv")))) + "hsw_decodern,hsw_p0,hsw_ssediv*20") + +(define_insn_reservation "hsw_sse_div_DF_load" 28 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "DF,V2DF,V4DF") + (eq_attr "type" "ssediv")))) + "hsw_decodern,(hsw_p23+hsw_p0),hsw_ssediv*20") + +(define_insn_reservation "hsw_sse_icvt" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "sseicvt"))) + "hsw_decodern,hsw_p1") + +(define_insn_reservation "hsw_sse_icvt_load" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (eq_attr "type" "sseicvt"))) + "hsw_decodern,hsw_p23+hsw_p1") + + +(define_insn_reservation "hsw_sse_icvt_SI" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (and (eq_attr "mode" "SI") + (eq_attr "type" "sseicvt")))) + "hsw_decodern,hsw_p1") + +(define_insn_reservation "hsw_sse_icvt_SI_load" 3 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "!none") + (and (eq_attr "mode" "SI") + (eq_attr "type" "sseicvt")))) + "hsw_decodern,hsw_p23+hsw_p1") + +(define_insn_reservation "hsw_sse_mov" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none") + (eq_attr "type" "ssemov"))) + "hsw_decodern,hsw_p015") + +(define_insn_reservation "hsw_sse_mov_load" 2 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "ssemov"))) + "hsw_decodern,hsw_p23") + +(define_insn_reservation "hsw_sse_mov_store" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (eq_attr "type" "ssemov"))) + "hsw_decodern,hsw_p4p7") + +(define_insn_reservation "hsw_insn" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "none,unknown") + (eq_attr "type" "alu,alu1,negnot,incdec,icmp,test,setcc,sseishft1,mmx,mmxcmp"))) + "hsw_decodern,hsw_p0156") + +(define_insn_reservation "hsw_insn_load" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "load") + (eq_attr "type" "alu,alu1,negnot,incdec,icmp,test,setcc,pop,sseishft1,mmx,mmxcmp"))) + "hsw_decodern,hsw_p23+hsw_p0156") + +(define_insn_reservation "hsw_insn_store" 1 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "store") + (eq_attr "type" "alu,alu1,negnot,incdec,icmp,test,setcc,sseishft1,mmx,mmxcmp"))) + "hsw_decodern,hsw_p0156+hsw_p4p7") + +;; read-modify-store instructions produce 4 uops so they have to be +;; decoded on hsw_decoder0 as well. +(define_insn_reservation "hsw_insn_both" 4 + (and (eq_attr "cpu" "haswell") + (and (eq_attr "memory" "both") + (eq_attr "type" "alu,alu1,negnot,incdec,icmp,test,setcc,pop,sseishft1,mmx,mmxcmp"))) + "hsw_decodern,hsw_p23+hsw_p0156+hsw_p4p7") diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c index d78f4e7..00e7006 100644 --- a/gcc/config/i386/i386.c +++ b/gcc/config/i386/i386.c @@ -3345,10 +3345,10 @@ ix86_option_override_internal (bool main_args_p, PTA_IVYBRIDGE}, {"core-avx-i", PROCESSOR_SANDYBRIDGE, CPU_NEHALEM, PTA_IVYBRIDGE}, - {"haswell", PROCESSOR_HASWELL, CPU_NEHALEM, PTA_HASWELL}, - {"core-avx2", PROCESSOR_HASWELL, CPU_NEHALEM, PTA_HASWELL}, - {"broadwell", PROCESSOR_HASWELL, CPU_NEHALEM, PTA_BROADWELL}, - {"skylake", PROCESSOR_HASWELL, CPU_NEHALEM, PTA_SKYLAKE}, + {"haswell", PROCESSOR_HASWELL, CPU_HASWELL, PTA_HASWELL}, + {"core-avx2", PROCESSOR_HASWELL, CPU_HASWELL, PTA_HASWELL}, + {"broadwell", PROCESSOR_HASWELL, CPU_HASWELL, PTA_BROADWELL}, + {"skylake", PROCESSOR_HASWELL, CPU_HASWELL, PTA_SKYLAKE}, {"bonnell", PROCESSOR_BONNELL, CPU_ATOM, PTA_BONNELL}, {"atom", PROCESSOR_BONNELL, CPU_ATOM, PTA_BONNELL}, {"silvermont", PROCESSOR_SILVERMONT, CPU_SLM, PTA_SILVERMONT}, @@ -51732,7 +51732,7 @@ ix86_reassociation_width (unsigned int, machine_mode mode) if (INTEGRAL_MODE_P (mode) && TARGET_REASSOC_INT_TO_PARALLEL) return 2; else if (FLOAT_MODE_P (mode) && TARGET_REASSOC_FP_TO_PARALLEL) - return 2; + return ((TARGET_64BIT && ix86_tune == PROCESSOR_HASWELL)? 4 : 2); else return 1; } diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md index 7017913..7808705 100644 --- a/gcc/config/i386/i386.md +++ b/gcc/config/i386/i386.md @@ -407,8 +407,8 @@ ;; Processor type. (define_attr "cpu" "none,pentium,pentiumpro,geode,k6,athlon,k8,core2,nehalem, - atom,slm,generic,amdfam10,bdver1,bdver2,bdver3,bdver4, - btver2" + atom,slm,haswell,generic,amdfam10,bdver1,bdver2,bdver3, + bdver4,btver2" (const (symbol_ref "ix86_schedule"))) ;; A basic instruction type. Refinements due to arguments to be @@ -1166,6 +1166,7 @@ (include "atom.md") (include "slm.md") (include "core2.md") +(include "haswell.md") ;; Operand and operator predicates and constraints -- 2.7.4