--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/gcc/config/i386/haswell.md	Fri Oct 27 22:46:09 2017 +0900
@@ -0,0 +1,618 @@
+;; Scheduling for Haswell and derived processors.
+;; Copyright (C) 2004-2017 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
+;; <http://www.gnu.org/licenses/>.  */
+
+;; 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")