FreeCalypso > hg > tcs211-l1-reconst

comparison STATUS @ 284:33b35965b4a0

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
STATUS: began analysis of the differences in the non-exact objects
author Mychaela Falconia <falcon@freecalypso.org>
date Tue, 14 Mar 2017 19:07:14 +0000
parents 974835ad720c
children 4d9af8de3973
comparison
equal deleted inserted replaced
283:974835ad720c 284:33b35965b4a0
1 Objects in l1_ext.lib: 1 Objects in l1_ext.lib:
2 2
3 l1_afunc.obj: not exact, needs review 3 l1_afunc.obj: not exact, see explanation below
4 l1_async.obj: perfect match 4 l1_async.obj: perfect match
5 l1_cmplx.obj: perfect match to 20070608 version except line number points 5 l1_cmplx.obj: perfect match to 20070608 version except line number points
6 l1_init.obj: perfect match 6 l1_init.obj: perfect match
7 l1_pwmgr.obj: perfect match 7 l1_pwmgr.obj: perfect match
8 l1_small.obj: perfect match 8 l1_small.obj: perfect match
10 l1_trace.obj: bits match perfectly, diff only in symbolic metadata 10 l1_trace.obj: bits match perfectly, diff only in symbolic metadata
11 11
12 l1_dyn_dwl_afunc.obj: perfect match 12 l1_dyn_dwl_afunc.obj: perfect match
13 l1_dyn_dwl_apihisr.obj: perfect match 13 l1_dyn_dwl_apihisr.obj: perfect match
14 l1_dyn_dwl_async.obj: perfect match 14 l1_dyn_dwl_async.obj: perfect match
15 l1_dyn_dwl_func.obj: not exact, needs review 15 l1_dyn_dwl_func.obj: not exact, see explanation below
16 l1_dyn_dwl_init.obj: sole diff is in the num_of_primitives oddity 16 l1_dyn_dwl_init.obj: sole diff is in the num_of_primitives oddity,
17 see explanation below
17 l1_dyn_dwl_sync.obj: perfect match 18 l1_dyn_dwl_sync.obj: perfect match
18 19
19 l1audio_abb.obj: perfect match 20 l1audio_abb.obj: perfect match
20 l1audio_afunc.obj: bits match perfectly, diff only in symbolic metadata 21 l1audio_afunc.obj: bits match perfectly, diff only in symbolic metadata
21 l1audio_async.obj: diffs only in register alloc and symbolic metadata 22 l1audio_async.obj: diffs only in register alloc and symbolic metadata
55 56
56 Objects in tpudrv.lib: 57 Objects in tpudrv.lib:
57 58
58 tpudrv.obj: perfect match 59 tpudrv.obj: perfect match
59 tpudrv12.obj: not exact, needs review 60 tpudrv12.obj: not exact, needs review
61
62 Detailed analysis of differences, i.e., cases where the reconstructed C code
63 compiles into an object that is not bit-identical to the original blob:
64
65 l1_afunc.obj:
66
67 The l1_afunc.c module from LoCosto is used without any changes:
68 it compiled right away after adding one constant to l1_const.h,
69 and the result of recompilation is identical to the original up until
70 the l1a_clip_txpwr() function, which is the last function in the module.
71
72 Disassembly of the original blob version of this function has been
73 analyzed and found to be identical in logic to the available C code,
74 but there must have been some change to the code expression that
75 results in different output from the compiler. In particular, the
76 symbolic info indicates the present of a local variable named power
77 in the original, but no such local var exists in the LoCosto version.
78 The TSM30 version was of no help as it is entirely different there.
79
80 See g23m/objdiff/l1_ext/l1_afunc.notes for further info.
81
82 l1_sync.obj: needs further review
83 l1_sync_intram.obj: ditto
84
85 dl1_com.obj: not exact, needs review
86
87 l1_dyn_dwl_func.obj:
88
89 The LoCosto version uses a function (implemented in l1_func.c in
90 the LoCosto code) named l1_memcpy_16bit() to download patch bits
91 into the DSP's API RAM; the function does what the name says, and
92 the logic of using 16-bit accesses when hitting this API RAM makes
93 perfect sense to me (Mychaela). The original TCS211 code apparently
94 used plain memcpy(), which the compiler turned into its C$MEMCPY call.
95
96 If one does a #define l1_memcpy_16bit memcpy in l1_dyn_dwl_func.c,
97 the resulting l1_dyn_dwl_func.obj is almost bit-identical to the
98 original blob: the only diff is in register allocation - the compiler's
99 register allocator picks a different choice of registers, but all
100 instructions still perfectly line up.
101
102 For production use of our reconstructed TCS211 L1 code, we have adopted
103 the version with l1_memcpy_16bit() - the function body itself has been
104 moved into l1_dyn_dwl_func.c so that l1_func.c remains a perfect match.
105
106 l1_dyn_dwl_init.obj:
107
108 In the TCS211 configuration without L1_GTT, there are a total of 5
109 signals (aka primitives aka L1 messages) which trigger a dynamic
110 DSP patch download. These 5 signals are enumerated in
111 signal_patch_array[] in l1_dyn_dwl_afunc.c (which has been reconstructed
112 to a perfect match), and their total count (which should be 5) goes into
113 l1a.dyn_dwnld.num_of_primitives, set in l1_dyn_dwnld_initialize_var() in
114 l1_dyn_dwl_init.c.
115
116 In the original blob version this num_of_primitives variable ends up
117 being set to 6: first set to 2, then incremented by 4. In the LoCosto
118 source this var is first set to 0 as they apparently have no patches
119 that are independent of compilation config options, and then incremented
120 for every enabled option.
121
122 In our reconstruction we have made the GPRS and AMR_SCH patches and the
123 primitives that trigger them mandatory, and conditionalized the E2 and
124 AMR_MMS patches on MELODY_E2. The GPRS and AMR_MMS critters are
125 preinstalled by the static patch, and get pushed out by AMR_SCH and E2,
126 respectively. Of the 5 primitives listed in signal_patch_array[] in
127 the original blob, the first 2 are always-present core ones, whereas
128 the last 3 are specific to E2. The AMR_MMS patch is reinstated upon
129 L1_BACK_MELODY_E2_UNLOAD_INSTRUMENT_CON.
130
131 Based on the above reasoning, our reconstructed version sets
132 l1a.dyn_dwnld.num_of_primitives to 5 in the full configuration: first
133 sets to 2, then increments by 3. If one makes a sans-MELODY_E2 build,
134 only the GPRS and AMR_SCH patches will be retained, and only the two
135 non-E2 primitives for entry into and exit from dedicated mode.
136
137 l1audio_async.obj:
138
139 The only bit diff is in register allocation - the compiler's register
140 allocator picks a different choice of registers, but all instructions
141 still perfectly line up. There is also a slight diff in the symbolic
142 metadata emitted for functions, which might be related to the diff in
143 the register allocation.
144
145 The root cause of these diffs could not be found, but there is no
146 functional difference as only some in-function temporary registers
147 have been interchanged.
148
149 l1audio_cust.obj:
150
151 The original blob version has these demo/test functions at the end of
152 the module:
153
154 audio_melo_e1_demo1_start()
155 audio_melo_e1_demo1_stop()
156 audio_melo_e1_demo2_start()
157 audio_melo_e1_demo2_stop()
158 audio_melo_e2_load_lsi()
159 audio_melo_e2_demo1_start()
160 audio_melo_e2_demo1_stop()
161 audio_melo_e2_demo2_start()
162 audio_melo_e2_demo2_stop()
163
164 These functions are dead code, i.e., they are not called or referenced
165 from anywhere else in the firmware, and the link succeeds without them
166 being present. These dead functions have been omitted from the
167 reconstructed version. The reconstruction is a perfect match otherwise.
168
169 tpudrv12.obj: early reconstruction, needs further review