Fix unified syntax in ARM inline assembly

GCC 4.9 always emits assembly with divided syntax. Setting unified
syntax in inline assembly causes the assembler to complain about
GCC's generated code, because the directive extends past the scope
of the inline asm. Fix this by setting divided mode at the end of
the inline assembly block.

The assembler directives are hidden behind macros because later
versions of GCC won't need this workaround: they can be told to
use the unified syntax with -masm-syntax-unified.

Change-Id: Ic09e729e5bbb6fd44d08dac348daf6f55c75d7d8

7 files changed, 28 insertions, 9 deletions

diff --git a/apps/plugins/mpegplayer/mpeg_misc.h b/apps/plugins/mpegplayer/mpeg_misc.h
index 68ee8cac3c..233b815493 100644
--- a/apps/plugins/mpegplayer/mpeg_misc.h
+++ b/apps/plugins/mpegplayer/mpeg_misc.h
@@ -53,13 +53,14 @@ enum state_enum
 #define CMP_3_CONST(_a, _b) \
     ({  int _x;                             \
         asm volatile (                      \
-            ".syntax unified          \n"   \
+            BEGIN_ARM_ASM_SYNTAX_UNIFIED    \
             "ldrb   %[x], [%[a], #0]  \n"   \
             "eors   %[x], %[x], %[b0] \n"   \
             "ldrbeq %[x], [%[a], #1]  \n"   \
             "eorseq %[x], %[x], %[b1] \n"   \
             "ldrbeq %[x], [%[a], #2]  \n"   \
             "eorseq %[x], %[x], %[b2] \n"   \
+            END_ARM_ASM_SYNTAX_UNIFIED      \
             : [x]"=&r"(_x)                  \
             : [a]"r"(_a),                   \
               [b0]"i"(((_b) >> 24) & 0xff), \
@@ -71,7 +72,7 @@ enum state_enum
 #define CMP_4_CONST(_a, _b) \
     ({  int _x;                             \
         asm volatile (                      \
-            ".syntax unified          \n"   \
+            BEGIN_ARM_ASM_SYNTAX_UNIFIED    \
             "ldrb   %[x], [%[a], #0]  \n"   \
             "eors   %[x], %[x], %[b0] \n"   \
             "ldrbeq %[x], [%[a], #1]  \n"   \
@@ -80,6 +81,7 @@ enum state_enum
             "eorseq %[x], %[x], %[b2] \n"   \
             "ldrbeq %[x], [%[a], #3]  \n"   \
             "eorseq %[x], %[x], %[b3] \n"   \
+            END_ARM_ASM_SYNTAX_UNIFIED      \
             : [x]"=&r"(_x)                  \
             : [a]"r"(_a),                   \
               [b0]"i"(((_b) >> 24) & 0xff), \
diff --git a/firmware/asm/arm/corelock.c b/firmware/asm/arm/corelock.c
index 07ec77a60e..a60299436f 100644
--- a/firmware/asm/arm/corelock.c
+++ b/firmware/asm/arm/corelock.c
@@ -61,7 +61,7 @@ int corelock_try_lock(struct corelock *cl)
 
     /* Relies on the fact that core IDs are complementary bitmasks (0x55,0xaa) */
     asm volatile (
-        ".syntax unified             \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
         "mov    r1, %[id]            \n" /* r1 = PROCESSOR_ID */
         "ldrb   r1, [r1]             \n"
         "strb   r1, [%[cl], r1, lsr #7] \n" /* cl->myl[core] = core */
@@ -74,6 +74,7 @@ int corelock_try_lock(struct corelock *cl)
         "ands   %[rv], %[rv], r1     \n"
         "strbeq %[rv], [%[cl], r1, lsr #7] \n" /* if not, cl->myl[core] = 0 */
     "1:                              \n" /* Done */
+        END_ARM_ASM_SYNTAX_UNIFIED
         : [rv] "=r"(rval)
         : [id] "i" (&PROCESSOR_ID), [cl] "r" (cl)
         : "r1","r2","cc"
diff --git a/firmware/asm/arm/thread.c b/firmware/asm/arm/thread.c
index bd9f950616..30df56e0d9 100644
--- a/firmware/asm/arm/thread.c
+++ b/firmware/asm/arm/thread.c
@@ -73,7 +73,7 @@ static inline void store_context(void* addr)
 static inline void load_context(const void* addr)
 {
     asm volatile(
-        ".syntax unified                \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
         "ldr     r0, [%0, #40]          \n" /* Load start pointer */
         "cmp     r0, #0                 \n" /* Check for NULL */
 
@@ -86,6 +86,7 @@ static inline void load_context(const void* addr)
 #endif
 
         "ldmia   %0, { r4-r11, sp, lr } \n" /* Load regs r4 to r14 from context */
+        END_ARM_ASM_SYNTAX_UNIFIED
         : : "r" (addr) : "r0" /* only! */
     );
 }
diff --git a/firmware/export/config.h b/firmware/export/config.h
index 19ee03b4c7..49cd6d610c 100644
--- a/firmware/export/config.h
+++ b/firmware/export/config.h
@@ -1005,6 +1005,16 @@ Lyre prototype 1 */
 #define ROCKBOX_STRICT_ALIGN 1
 #endif
 
+/*
+ * These macros are for switching on unified syntax in inline assembly.
+ * Older versions of GCC emit assembly in divided syntax with no option
+ * to enable unified syntax.
+ *
+ * FIXME: This needs to be looked at after the toolchain bump
+ */
+#define BEGIN_ARM_ASM_SYNTAX_UNIFIED ".syntax unified\n"
+#define END_ARM_ASM_SYNTAX_UNIFIED   ".syntax divided\n"
+
 #if defined(CPU_ARM) && defined(__ASSEMBLER__)
 .syntax unified
 /* ARMv4T doesn't switch the T bit when popping pc directly, we must use BX */
diff --git a/firmware/target/arm/pcm-telechips.c b/firmware/target/arm/pcm-telechips.c
index 747765d8fb..45044bc664 100644
--- a/firmware/target/arm/pcm-telechips.c
+++ b/firmware/target/arm/pcm-telechips.c
@@ -218,7 +218,7 @@ void fiq_handler(void)
      * r0-r3 and r12 is a working register.
      */
     asm volatile (
-        ".syntax unified             \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
         "sub     lr, lr, #4          \n"
         "stmfd   sp!, { r0-r3, lr }  \n" /* stack scratch regs and lr */
         "mov     r14, #0             \n" /* Was the callback called? */
@@ -274,6 +274,7 @@ void fiq_handler(void)
         "bhi     .fill_fifo          \n" /* not stop and enough? refill */
         "ldmfd   sp!, { r0-r3, pc }^ \n" /* exit */
         ".ltorg                      \n"
+        END_ARM_ASM_SYNTAX_UNIFIED
         : : "i"(PCM_DMAST_OK), "i"(PCM_DMAST_STARTED)
     );
 }
diff --git a/firmware/target/arm/pp/pcm-pp.c b/firmware/target/arm/pp/pcm-pp.c
index fd798f0506..94b1c5ae10 100644
--- a/firmware/target/arm/pp/pcm-pp.c
+++ b/firmware/target/arm/pp/pcm-pp.c
@@ -327,7 +327,7 @@ void fiq_playback(void)
      */
     asm volatile (
     /* No external calls */
-        ".syntax unified              \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
         "sub     lr, lr, #4           \n" /* Prepare return address */
         "stmfd   sp!, { lr }          \n" /* stack lr so we can use it */
         "ldr     r12, =0xcf001040     \n" /* Some magic from iPodLinux ... */
@@ -395,6 +395,7 @@ void fiq_playback(void)
         "bne     3b                   \n" /* no? -> go return */
         "b       2b                   \n" /* yes -> get even more */
         ".ltorg                       \n"
+        END_ARM_ASM_SYNTAX_UNIFIED
         : /* These must only be integers! No regs */
         : "i"(PCM_DMAST_OK), "i"(PCM_DMAST_STARTED));
 }
diff --git a/lib/rbcodec/codecs/demac/libdemac/vector_math16_armv6.h b/lib/rbcodec/codecs/demac/libdemac/vector_math16_armv6.h
index 1da090efbb..ad5eed60fb 100644
--- a/lib/rbcodec/codecs/demac/libdemac/vector_math16_armv6.h
+++ b/lib/rbcodec/codecs/demac/libdemac/vector_math16_armv6.h
@@ -45,7 +45,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2)
 #endif
 
     asm volatile (
-        ".syntax unified                 \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
 #if ORDER > 32
         "mov     %[res], #0              \n"
 #endif
@@ -186,6 +186,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2)
 #endif
 
     "99:                                 \n"
+        END_ARM_ASM_SYNTAX_UNIFIED
         : /* outputs */
 #if ORDER > 32
         [cnt]"+r"(cnt),
@@ -215,7 +216,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2)
 #endif
 
     asm volatile (
-        ".syntax unified                 \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
 #if ORDER > 32
         "mov     %[res], #0              \n"
 #endif
@@ -356,6 +357,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2)
 #endif
 
     "99:                                 \n"
+        END_ARM_ASM_SYNTAX_UNIFIED
         : /* outputs */
 #if ORDER > 32
         [cnt]"+r"(cnt),
@@ -383,7 +385,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2)
 #endif
 
     asm volatile (
-        ".syntax unified                 \n"
+        BEGIN_ARM_ASM_SYNTAX_UNIFIED
 #if ORDER > 32
         "mov     %[res], #0              \n"
 #endif
@@ -477,6 +479,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2)
 #endif
 
     "99:                                 \n"
+        END_ARM_ASM_SYNTAX_UNIFIED
         : /* outputs */
 #if ORDER > 32
         [cnt]"+r"(cnt),