From b1caf4a07d888b257b984c25b52548dabe0ff0ab Mon Sep 17 00:00:00 2001
From: Andrew Mahone <andrew.mahone@gmail.com>
Date: Thu, 4 Feb 2010 08:45:38 +0000
Subject: Use all available codec iram for reciprocal table in APE codec on
 ARMv4. Done by linking first with the table empty to determine free space,
 then sizing table to fill it.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@24507 a1c6a512-1295-4272-9138-f99709370657
---
 apps/codecs/codecs.make                     | 10 ++++++-
 apps/codecs/demac/libdemac.make             | 17 ++++++++++++
 apps/codecs/demac/libdemac/udiv32_arm-pre.S | 25 +++++++++++++++++
 apps/codecs/demac/libdemac/udiv32_arm.S     | 42 ++++++++++++++++++++---------
 4 files changed, 80 insertions(+), 14 deletions(-)
 create mode 100644 apps/codecs/demac/libdemac/udiv32_arm-pre.S

(limited to 'apps/codecs')

diff --git a/apps/codecs/codecs.make b/apps/codecs/codecs.make
index 9e1f909462..0c3d2fade0 100644
--- a/apps/codecs/codecs.make
+++ b/apps/codecs/codecs.make
@@ -59,7 +59,7 @@ CODECLIBS := $(DEMACLIB) $(A52LIB) $(ALACLIB) $(ASAPLIB) \
 	$(ATRACLIB) \
 	$(CODECLIB)
 
-$(CODECS): $(CODEC_CRT0) $(CODECLINK_LDS) 
+$(CODECS): $(CODEC_CRT0) $(CODECLINK_LDS)
 
 $(CODECLINK_LDS): $(CODEC_LDS) $(CONFIGFILE)
 	$(call PRINTS,PP $(@F))
@@ -78,6 +78,7 @@ $(CODECDIR)/wavpack.codec : $(CODECDIR)/libwavpack.a
 $(CODECDIR)/alac.codec : $(CODECDIR)/libalac.a $(CODECDIR)/libm4a.a 
 $(CODECDIR)/aac.codec : $(CODECDIR)/libfaad.a $(CODECDIR)/libm4a.a
 $(CODECDIR)/shorten.codec : $(CODECDIR)/libffmpegFLAC.a
+$(CODECDIR)/ape-pre.map : $(CODECDIR)/libdemac-pre.a
 $(CODECDIR)/ape.codec : $(CODECDIR)/libdemac.a
 $(CODECDIR)/wma.codec : $(CODECDIR)/libwma.a
 $(CODECDIR)/wavpack_enc.codec: $(CODECDIR)/libwavpack.a
@@ -113,6 +114,13 @@ else
  CODECFLAGS += -UDEBUG -DNDEBUG
 endif
 
+$(CODECDIR)/%-pre.map: $(CODEC_CRT0) $(CODECLINK_LDS) $(CODECDIR)/%.o $(CODECLIB)
+	$(call PRINTS,LD $(@F))$(CC) $(CODECFLAGS) -o $(CODECDIR)/$*-pre.elf \
+		$(filter %.o, $^) \
+		$(filter-out $(CODECLIB),$(filter %.a, $+)) \
+		$(CODECLIB) \
+		-lgcc $(subst .map,-pre.map,$(CODECLDFLAGS))
+
 $(CODECDIR)/%.codec: $(CODECDIR)/%.o
 	$(call PRINTS,LD $(@F))$(CC) $(CODECFLAGS) -o $(CODECDIR)/$*.elf \
 		$(filter %.o, $^) \
diff --git a/apps/codecs/demac/libdemac.make b/apps/codecs/demac/libdemac.make
index 5026fb1d95..dcbd35c6b1 100644
--- a/apps/codecs/demac/libdemac.make
+++ b/apps/codecs/demac/libdemac.make
@@ -12,6 +12,15 @@ DEMACLIB := $(CODECDIR)/libdemac.a
 DEMACLIB_SRC := $(call preprocess, $(APPSDIR)/codecs/demac/libdemac/SOURCES)
 DEMACLIB_OBJ := $(call c2obj, $(DEMACLIB_SRC))
 OTHER_SRC += $(DEMACLIB_SRC)
+ifeq ($(CPU),arm)
+OTHER_SRC += $(APPSDIR)/codecs/demac/libdemac/udiv32_arm-pre.S
+endif
+DEMACLIB_PRE := $(subst .a,-pre.a,$(DEMACLIB))
+DEMACLIB_OBJ_PRE := $(subst udiv32_arm.o,udiv32_arm-pre.o,$(DEMACLIB_OBJ))
+
+$(DEMACLIB_PRE): $(DEMACLIB_OBJ_PRE)
+	$(SILENT)$(shell rm -f $@)
+	$(call PRINTS,AR $(@F))$(AR) rcs $@ $^ >/dev/null
 
 $(DEMACLIB): $(DEMACLIB_OBJ)
 	$(SILENT)$(shell rm -f $@)
@@ -20,6 +29,14 @@ $(DEMACLIB): $(DEMACLIB_OBJ)
 DEMACFLAGS = $(filter-out -O%,$(CODECFLAGS))
 DEMACFLAGS += -O3
 
+$(CODECDIR)/ape_free_iram.h: $(CODECDIR)/ape-pre.map
+	$(call PRINTS,GEN $(@F))perl -an \
+		-e 'if(/^PLUGIN_IRAM/){$$istart=hex($$F[1]);$$ilen=hex($$F[2])}' \
+		-e 'if(/iend = /){$$iend=hex($$F[0]);}' \
+		-e '}{if($$ilen){print"#define FREE_IRAM ".($$ilen+$$istart-$$iend)."\n";}' \
+		$(CODECDIR)/ape-pre.map \
+		> $@
+
 $(CODECDIR)/demac/%.o: $(ROOTDIR)/apps/codecs/demac/%.c
 	$(SILENT)mkdir -p $(dir $@)
 	$(call PRINTS,CC $(subst $(ROOTDIR)/,,$<))$(CC) $(DEMACFLAGS) -c $< -o $@
diff --git a/apps/codecs/demac/libdemac/udiv32_arm-pre.S b/apps/codecs/demac/libdemac/udiv32_arm-pre.S
new file mode 100644
index 0000000000..882d9b6f50
--- /dev/null
+++ b/apps/codecs/demac/libdemac/udiv32_arm-pre.S
@@ -0,0 +1,25 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2010 by Andrew Mahone
+ *
+ * Wrapper for udiv32_arm.S to test available IRAM by pre-linking the codec.
+ *
+ * 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.
+ *
+ * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
+ * KIND, either express or implied.
+ *
+ ****************************************************************************/
+
+#define APE_PRE
+#include "udiv32_arm.S"
\ No newline at end of file
diff --git a/apps/codecs/demac/libdemac/udiv32_arm.S b/apps/codecs/demac/libdemac/udiv32_arm.S
index 4492492d30..939fce17d4 100644
--- a/apps/codecs/demac/libdemac/udiv32_arm.S
+++ b/apps/codecs/demac/libdemac/udiv32_arm.S
@@ -29,6 +29,13 @@
  ****************************************************************************/
 
 #include "config.h"
+/* On targets with codec iram, a header file will be generated after an initial
+   link of the APE codec, stating the amount of IRAM remaining for use by the
+   reciprocal lookup table. */
+#if !defined(APE_PRE) && defined(USE_IRAM) && ARM_ARCH < 5
+#include "apps/codecs/ape_free_iram.h"
+#endif
+
 /* Codecs should not normally do this, but we need to check a macro, and
  * codecs.h would confuse the assembler. */
 
@@ -88,20 +95,29 @@
     eor     \quotient, \result, \remainder, lsl \bits
 .endm
 
-#ifdef CPU_PP
-#if CONFIG_CPU == PP5020
-.set recip_max, 8384
-#elif CONFIG_CPU == PP5002
-.set recip_max, 4608
+#ifndef FREE_IRAM
+.set recip_max, 2
 #else
-.set recip_max, 16384
-#endif
-#elif CONFIG_CPU == AS3525
-.set recip_max, 42752
-#elif CONFIG_CPU == S5L8701
-.set recip_max, 12800
-#elif CONFIG_CPU == S5L8700
-.set recip_max, 9088
+/* Each table entry is one word. Since a compare is done against the maximum
+   entry as an immediate, the maximum entry must be a valid ARM immediate,
+   which means a byte shifted by an even number of places. */
+.set recip_max, 2 + FREE_IRAM / 4
+.set recip_max_tmp, recip_max >> 8
+.set recip_mask_shift, 0
+.set tmp_shift, 16
+.rept 5
+    .if recip_max_tmp >> tmp_shift
+        .set recip_max_tmp, recip_max_tmp >> tmp_shift
+        .set recip_mask_shift, recip_mask_shift + tmp_shift
+    .endif
+    .set tmp_shift, tmp_shift >> 1
+.endr
+.if recip_max_tmp
+    .set recip_mask_shift, recip_mask_shift + 1
+.endif
+.set recip_mask_shift, (recip_mask_shift + 1) & 62
+.set recip_max, recip_max & (255 << recip_mask_shift)
+//.set recip_max, 2
 #endif
 
 udiv32_arm:
-- 
cgit v1.2.3