This is an automated email from the ASF dual-hosted git repository.
pkarashchenko pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/nuttx.git
commit 9ab97df0a57d92caf509824c9bc15309a71bc242
Author: chao an <anc...@xiaomi.com>
AuthorDate: Fri Dec 2 02:20:09 2022 +0800
arm/cortex-[a|r]/thumb: force assembler files to be interpreted as Thumb
code
The option '-mthumb' is only valid for C source files and it is not passed
to the assembler.
If 'thumb' is not considered in some assembly projects, the system will
generate
'undefined instructions' when running incompatible instruction:
arm_undefinedinsn: Undefined instruction at 0x380cfc98
This assembly file should be compiled with .thumb but it doesn't:
380cfc90 <hobot_i8_i32_gemm_nn_m4_n8_neon>:
380cfc90: e92d4ff0 push {r4, r5, r6, r7, r8, r9, sl, fp, lr}
380cfc94: ed2d8b10 vpush {d8-d15}
380cfc98: e59d4064 ldr r4, [sp, #100] ; 0x64 <-- Undefined instruction
380cfc9c: e59d5068 ldr r5, [sp, #104] ; 0x68
380cfca0: e59d606c ldr r6, [sp, #108] ; 0x6c
380cfca4: e59d7070 ldr r7, [sp, #112] ; 0x70
380cfca8: e1a08120 lsr r8, r0, #2
380cfcac: e1a091a1 lsr r9, r1, #3
380cfcb0: e1a0a122 lsr sl, r2, #2
After enable thumb:
.syntax unified
.thumb
or
-Wa,-mthumb
.Lhobot_i8_i32_gemm_nn_m4_n8_neon:
38001100: e92d 4ff0 stmdb sp!, {r4, r5, r6, r7, r8, r9, sl, fp, lr}
38001104: ed2d 8b10 vpush {d8-d15}
38001108: 9c19 ldr r4, [sp, #100] ; 0x64
3800110a: 9d1a ldr r5, [sp, #104] ; 0x68
3800110c: 9e1b ldr r6, [sp, #108] ; 0x6c
3800110e: 9f1c ldr r7, [sp, #112] ; 0x70
38001110: ea4f 0890 mov.w r8, r0, lsr #2
38001114: ea4f 09d1 mov.w r9, r1, lsr #3
38001118: ea4f 0a92 mov.w sl, r2, lsr #2
This commit will enable the thumb option of the assembly file by default,
so that when compiling the assembly file, the machine code and the system
will be in a consistent state.
----------------------------------------------------------------
https://gcc.gnu.org/onlinedocs/gcc-4.5.2/gcc/ARM-Options.html
GCC Manual:
-mthumb
Generate code for the Thumb instruction set. The default is to use the
32-bit ARM
instruction set. This option automatically enables either 16-bit Thumb-1
or mixed 16/32-bit
Thumb-2 instructions based on the -mcpu=name and -march=name options.
** This option is not passed to the assembler. **
** If you want to force assembler files to be interpreted as Thumb code,
either add a `.thumb' directive to the source or pass the -mthumb
option
directly to the assembler by prefixing it with -Wa. **
Signed-off-by: chao an <anc...@xiaomi.com>
---
arch/arm/src/common/Toolchain.defs | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/arch/arm/src/common/Toolchain.defs
b/arch/arm/src/common/Toolchain.defs
index c5c762bced..fa3624d2b0 100644
--- a/arch/arm/src/common/Toolchain.defs
+++ b/arch/arm/src/common/Toolchain.defs
@@ -72,6 +72,14 @@ endif
ifeq ($(CONFIG_ARM_THUMB),y)
ARCHOPTIMIZATION += -mthumb
+
+ # GCC Manual:
+ # -mthumb
+ # ... If you want to force assembler files to be interpreted as Thumb
+ # code, either add a `.thumb' directive to the source or pass the
+ # -mthumb option directly to the assembler by prefixing it with -Wa.
+
+ ARCHOPTIMIZATION += -Wa,-mthumb
endif
ifeq ($(CONFIG_UNWINDER_ARM),y)
