jumper, revert to generating .S files

I went with the unified-in-one-.cpp approach mostly to make it easy to
roll out SkJumper.  I no longer see any difficultly rolling out the
assembly files, and it's possible the unified .cpp approach just makes
things harder.

Let's see if it's any easier to get Chrome's official build to work with
normal assembly files.  It's not going to be a problem to roll out.

This is a partial revert of https://skia-review.googlesource.com/c/9336.

CQ_INCLUDE_TRYBOTS=skia.primary:Test-Win2k8-MSVC-GCE-CPU-AVX2-x86_64-Debug,Test-Mac-Clang-MacMini6.2-CPU-AVX-x86_64-Debug,Test-Ubuntu-Clang-GCE-CPU-AVX2-x86_64-Debug,Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Debug

Change-Id: Idfdbd2d322452b44bc0adaf6dc299cc7649bc51e
Reviewed-on: https://skia-review.googlesource.com/10561
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>

BUILD.gn

@@ -518,9 +518,13 @@ optional("jumper") {
   public_defines = [ "SK_JUMPER" ]
   sources = [
     "src/jumper/SkJumper.cpp",
-    "src/jumper/SkJumper_generated.cpp",
     "src/jumper/SkJumper_stages.cpp",
   ]
+  if (is_win && target_cpu == "x64") {
+    sources += [ "src/jumper/SkJumper_generated_win.S" ]
+  } else if (!is_win) {
+    sources += [ "src/jumper/SkJumper_generated.S" ]
+  }
 }
 
 optional("typeface_freetype") {

src/jumper/SkJumper.cpp

@@ -91,8 +91,12 @@ static K kConstants = {
 // We'll only ever call start_pipeline(), which then chains into the rest for us.
 using StageFn = void(void);
 
-// TODO: maybe don't need this wrapper anymore.
-#define ASM(name, suffix) sk_##name##_##suffix
+// Some platforms expect C "name" maps to asm "_name", others to "name".
+#if defined(__APPLE__)
+    #define ASM(name, suffix)  sk_##name##_##suffix
+#else
+    #define ASM(name, suffix) _sk_##name##_##suffix
+#endif
 
 extern "C" {
 

src/jumper/build_stages.py

@@ -71,7 +71,15 @@ subprocess.check_call(clang + cflags + vfp4 +
                       ['-c', 'src/jumper/SkJumper_stages.cpp'] +
                       ['-o', 'vfp4.o'])
 
-def parse_object_file(dot_o, array_type, target=None):
+def parse_object_file(dot_o, directive, target=None):
+  globl, label, comment = '.globl', ':', '// '
+  if 'win' in dot_o:
+    globl, label, comment = 'PUBLIC', ' LABEL PROC', '; '
+
+  dehex = lambda h: '0x'+h
+  if directive != '.long':
+    dehex = lambda h: str(int(h,16))
+
   cmd = [objdump]
   if target:
     cmd += ['--target', target]
@@ -84,7 +92,6 @@ def parse_object_file(dot_o, array_type, target=None):
       assert snippet not in section_headers
 
   # Ok.  Let's disassemble.
-  active = False
   disassemble = ['-d', '--insn-width=10', dot_o]
   for line in subprocess.check_output(cmd + disassemble).split('\n'):
     line = line.strip()
@@ -95,11 +102,9 @@ def parse_object_file(dot_o, array_type, target=None):
     # E.g. 00000000000003a4 <_load_f16>:
     m = re.match('''[0-9a-f]+ <_?(.*)>:''', line)
     if m:
-      if active:
-        print '};'
       print
-      print 'CODE const', array_type, m.group(1) + '[] = {'
-      active = True
+      print globl + ' _' + m.group(1)
+      print '_' + m.group(1) + label
       continue
 
     columns = line.split('\t')
@@ -113,54 +118,50 @@ def parse_object_file(dot_o, array_type, target=None):
         inst, args = columns[2].split(' ', 1)
     code, inst, args = code.strip(), inst.strip(), args.strip()
 
-    dehex = lambda x: '0x'+x
-    if array_type == 'uint8_t':
-      dehex = lambda x: str(int(x, 16))
+    hexed = ','.join(dehex(x) for x in code.split(' '))
+    print '  ' + directive + '  ' + hexed + ' '*(36-len(hexed)) + \
+          comment + inst + (' '*(14-len(inst)) + args if args else '')
 
-    hexed = ''.join(dehex(x) + ',' for x in code.split(' '))
-    print '  ' + hexed + ' '*(40-len(hexed)) + \
-          '//' + inst  + (' '*(14-len(inst)) + args if args else '')
-  print '};'
+sys.stdout = open('src/jumper/SkJumper_generated.S', 'w')
 
-sys.stdout = open('src/jumper/SkJumper_generated.cpp', 'w')
+print '''# Copyright 2017 Google Inc.
+#
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
 
-print '''/*
- * Copyright 2017 Google Inc.
- *
- * Use of this source code is governed by a BSD-style license that can be
- * found in the LICENSE file.
- */
-
-// This file is generated semi-automatically with this command:
-//   $ src/jumper/build_stages.py
-
-#include <stdint.h>
-
-#if defined(_MSC_VER)
-    #pragma section("code", read,execute)
-    #define CODE extern "C" __declspec(allocate("code"))
-#elif defined(__MACH__)
-    #define CODE extern "C" __attribute__((section("__TEXT,__text")))
-#else
-    #define CODE extern "C" __attribute__((section(".text")))
-#endif
+# This file is generated semi-automatically with this command:
+#   $ src/jumper/build_stages.py
 '''
+print '.text'
 print '#if defined(__aarch64__)'
-parse_object_file('aarch64.o', 'uint32_t')
+print '.balign 4'
+parse_object_file('aarch64.o', '.long')
 
 print '#elif defined(__arm__)'
-parse_object_file('vfp4.o', 'uint32_t', target='elf32-littlearm')
+print '.balign 4'
+parse_object_file('vfp4.o', '.long', target='elf32-littlearm')
 
 print '#elif defined(__x86_64__)'
-parse_object_file('hsw.o',   'uint8_t')
-parse_object_file('avx.o',   'uint8_t')
-parse_object_file('sse41.o', 'uint8_t')
-parse_object_file('sse2.o',  'uint8_t')
-
-print '#elif defined(_M_X64)'
-parse_object_file('win_hsw.o',   'uint8_t')
-parse_object_file('win_avx.o',   'uint8_t')
-parse_object_file('win_sse41.o', 'uint8_t')
-parse_object_file('win_sse2.o',  'uint8_t')
+parse_object_file('hsw.o',   '.byte')
+parse_object_file('avx.o',   '.byte')
+parse_object_file('sse41.o', '.byte')
+parse_object_file('sse2.o',  '.byte')
 
 print '#endif'
+
+sys.stdout = open('src/jumper/SkJumper_generated_win.S', 'w')
+print '''; Copyright 2017 Google Inc.
+;
+; Use of this source code is governed by a BSD-style license that can be
+; found in the LICENSE file.
+
+; This file is generated semi-automatically with this command:
+;   $ src/jumper/build_stages.py
+'''
+
+print '_text SEGMENT'
+parse_object_file('win_hsw.o',   'DB')
+parse_object_file('win_avx.o',   'DB')
+parse_object_file('win_sse41.o', 'DB')
+parse_object_file('win_sse2.o',  'DB')
+print 'END'