starting to commit gridflow 0.8.0 ...

if you know how to use "cvs import" please mail me and i'll use it for 0.8.1

svn path=/trunk/; revision=3646

1 files changed, 225 insertions, 0 deletions

diff --git a/externals/gridflow/cpu/mmx.rb b/externals/gridflow/cpu/mmx.rb
new file mode 100644
index 00000000..1a3b15d3
--- /dev/null
+++ b/externals/gridflow/cpu/mmx.rb
@@ -0,0 +1,225 @@
+=begin
+	$Id: mmx.rb,v 1.1 2005-10-04 02:02:14 matju Exp $
+
+	GridFlow
+	Copyright (c) 2001,2002,2003,2004 by Mathieu Bouchard
+
+	This program 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 2
+	of the License, or (at your option) any later version.
+
+	See file ../COPYING for further informations on licensing terms.
+
+	This program 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 this program; if not, write to the Free Software
+	Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+=end
+
+STDOUT.reopen ARGV[0], "w"
+$loader = File.open ARGV[1], "w"
+$count = 0
+$lines = 0
+
+puts "; generated by/for GridFlow 0.8.0"
+$loader.puts "#include \"../base/grid.h.fcs\"\nextern \"C\" {"
+
+# this class is not really used yet (only self.make)
+class AsmFunction
+	def initialize(name)
+		@name = name
+		@label_count = 1
+	end
+	def self.make(name)
+		puts "", "GLOBAL #{name}", "#{name}:"
+		puts "push ebp", "mov ebp,esp", "push esi", "push edi"
+		yield AsmFunction.new(name)
+		puts "pop edi", "pop esi", "leave", "ret", ""
+	end
+	def make_until(*ops)
+		a = @label_count
+		b = @label_count+1
+		@label_count+=2
+		ops[-1]<<" #{@name}_#{b}"
+		puts "#{@name}_#{a}: ", *ops
+		yield
+		puts "jmp #{@name}_#{a}"
+		puts "#{@name}_#{b}:"
+	end
+end
+
+$sizeof = {
+	:uint8 => 1,
+	:int16 => 2,
+	:int32 => 4,
+	:int64 => 8,
+	:float32 => 4,
+	:float64 => 8,
+}
+
+$accum = {
+	:uint8 => "al",
+	:int16 => "ax",
+	:int32 => "eax",
+}
+
+$asm_type = {
+	:uint8 => "byte",
+	:int16 => "word",
+	:int32 => "dword",
+	:int64 => "qword",
+}
+
+# in the following, the opcode "_" means no such thing seems available.
+# also >> for x86 ought to be shr in the uint8 case.
+# btw, i got all of the MMX information from the NASM manual, Appendix B.
+$opcodes = {
+#                     [--GF--|--x86--|--mmx-et-al----------------------------------------]
+#                     [      |       |-uint8-|-int16-|-int32-|-int64-|-float32-|-float64-]
+	:add     => %w[ +      add    paddb   paddw   paddd   paddq                      ],
+	:sub     => %w[ -      sub    psubb   psubw   psubd   psubq                      ],
+	:and     => %w[ &      and    pand    pand    pand    pand                       ],
+	:xor     => %w[ ^      xor    pxor    pxor    pxor    pxor                       ],
+	:or      => %w[ |      or     por     por     por     por                        ],
+#	:max     => %w[ max    _      pmaxub  pmaxsw  _       _                          ], # not plain MMX !!! (req.Katmai)
+#	:min     => %w[ min    _      pminub  pminsw  _       _                          ], # not plain MMX !!! (req.Katmai)
+#	:eq      => %w[ ==     _      pcmpeqb pcmpeqw pcmpeqd _                          ],
+#	:gt      => %w[ >      _      pcmpgtb pcmpgtw pcmpgtd _                          ],
+#	:shl     => %w[ <<     shl    _       psllw   pslld   psllq                      ], # noncommutative
+#	:shr     => %w[ >>     sar    _       psraw   psrad   _                          ], # noncommutative
+#	:clipadd => %w[ clip+  _      paddusb paddsw  _       _                          ], # future use
+#	:clipsub => %w[ clip-  _      psubusb psubsw  _       _                          ], # future use
+#	:andnot  => %w[ &not   _      pandn   pandn   pandn   pandn                      ], # not planned
+}
+
+$opcodes.each {|k,op|
+	op.map! {|x| if x=="_" then nil else x end }
+	STDERR.puts op.inspect
+}
+
+$decls = ""
+$install = ""
+
+def make_fun_map(op,type)
+	s="mmx_#{type}_map_#{op}"
+	size = $sizeof[type]
+	accum = $accum[type]
+	sym = $opcodes[op][0]
+	opcode = $opcodes[op][1]
+	mopcode = $opcodes[op][size+(size<4 ? 1 : 0)]
+	return if not mopcode
+	AsmFunction.make(s) {|a|
+		puts "mov ecx,[ebp+8]", "mov esi,[ebp+12]", "mov eax,[ebp+16]"
+		puts "mov dx,ax", "shl eax,8", "mov al,dl" if size==1
+		puts "mov edx,eax", "shl eax,16", "mov ax,dx" if size<=2
+		puts "push eax", "push eax", "movq mm7,[esp]", "add esp,8"
+		foo = proc {|n|
+			a.make_until("cmp ecx,#{8/size*n}","jb near") {
+				0.step(n,4) {|k|
+				nn=[n-k,4].min
+				o=(0..3).map{|x| 8*(x+k) }
+				for i in 0...nn do puts "movq mm#{i},[esi+#{o[i]}]" end
+				for i in 0...nn do puts "#{mopcode} mm#{i},mm7" end
+				for i in 0...nn do puts "movq [esi+#{o[i]}],mm#{i}" end
+				}
+				puts "lea esi,[esi+#{8*n}]", "lea ecx,[ecx-#{8/size*n}]"
+			}
+		}
+		foo.call 4
+		foo.call 1
+		a.make_until("test ecx,ecx", "jz") {
+			puts "#{opcode} #{$asm_type[type]} [esi],#{accum}", "lea esi,[esi+#{size}]"
+			puts "dec ecx"
+		}
+		puts "emms"
+	}
+	$decls << "void #{s}(int,#{type}*,#{type});\n"
+	$install << "FIX2PTR(Numop,rb_hash_aref(op_dict,SYM(#{sym})))"
+	$install << "->on_#{type}.op_map = #{s};\n"
+	$count += 1
+end
+
+def make_fun_zip(op,type)
+s="mmx_#{type}_zip_#{op}"
+	size = $sizeof[type]
+	accum = $accum[type]
+	sym = $opcodes[op][0]
+	opcode = $opcodes[op][1]
+	mopcode = $opcodes[op][size+(size<4 ? 1 : 0)]
+	return if not mopcode
+	AsmFunction.make(s) {|a|
+		puts "mov ecx,[ebp+8]",  "mov edi,[ebp+12]",
+		     "mov esi,[ebp+16]"#, "mov ebx,[ebp+20]"
+		foo = proc {|n|
+			a.make_until("cmp ecx,#{8/size*n}","jb near") {
+				0.step(n,4) {|k|
+				nn=[n-k,4].min
+				o=(0..3).map{|x| 8*(x+k) }
+				for i in 0...nn do puts "movq mm#{i},[edi+#{o[i]}]" end
+				for i in 0...nn do puts "movq mm#{i+4},[esi+#{o[i]}]" end
+				for i in 0...nn do puts "#{mopcode} mm#{i},mm#{i+4}" end
+				for i in 0...nn do puts "movq [edi+#{o[i]}],mm#{i}" end
+				}
+				#for i in 0...n do puts "movq [ebx+#{8*i}],mm#{i}" end
+				puts "lea edi,[edi+#{8*n}]"
+				puts "lea esi,[esi+#{8*n}]"
+				#puts "lea ebx,[ebx+#{8*n}]"
+				puts "lea ecx,[ecx-#{8/size*n}]"
+			}
+		}
+		foo.call 4
+		foo.call 1
+		a.make_until("test ecx,ecx", "jz") {
+			# requires commutativity ??? fails with shl, shr
+			puts "mov #{accum},[esi]"
+			puts "#{opcode} #{$asm_type[type]} [edi],#{accum}"
+			#puts "mov #{accum},[edi]"
+			#puts "#{opcode} #{accum},[esi]"
+			#puts "mov [ebx],#{accum}"
+			puts "lea edi,[edi+#{size}]"
+			puts "lea esi,[esi+#{size}]"
+			#puts "lea ebx,[ebx+#{size}]"
+			puts "dec ecx"
+		}
+		puts "emms"
+	}
+	#$decls << "void #{s}(int,#{type}*,#{type}*,#{type}*);\n"
+	$decls << "void #{s}(int,#{type}*,#{type}*);\n"
+	$install << "FIX2PTR(Numop,rb_hash_aref(op_dict,SYM(#{sym})))"
+	$install << "->on_#{type}.op_zip = #{s};\n"
+	$count += 1
+end
+
+for op in $opcodes.keys do
+	for type in [:uint8, :int16#, :int32
+	] do
+		make_fun_map(op,type)
+		make_fun_zip(op,type)
+	end
+end
+
+$loader.puts $decls
+$loader.puts %`
+}; /* extern */
+#include <stdlib.h>
+void startup_mmx_loader () {/*bogus*/}
+void startup_mmx () {
+	if (getenv("NO_MMX")) return;
+	if (EVAL(\"GridFlow.bridge_name\")!=Qnil) gfpost(\"startup_cpu: using MMX optimisations\");
+	#{$install}
+}`
+
+STDERR.puts "automatically generated #{$count} MMX asm functions"
+
+=begin notes:
+CPUID has a bit for detecting MMX
+PACKSSDW PACKSSWB PACKUSWB = saturation-casting
+PCMPxx: Compare Packed Integers
+PMULHW, PMULLW: Multiply Packed _unsigned_ 16-bit Integers, and Store
+PUNPCKxxx: Unpack and Interleave Data
+=end