[AVR] llvm-objdump should automatically select the instruction set

[benshi001]

Given a simple test program:

int adiw(int a) {
	return a + 51;
}

Build it with clang a.c --target=avr -mmcu=avr2 -c -O3, and then disassemble it with llvm-objdump -d a.o, we see

a.o:	file format elf32-avr

Disassembly of section .text:

00000000 <adiw>:
       0: c3 96 08 95  	<unknown>

We must specify --mcpu=avr2 to llvm-objdump explicitly to make adiw correctly decoded.

a.o:	file format elf32-avr

Disassembly of section .text:

00000000 <adiw>:
       0: c3 96        	adiw	r24, 0x33
       2: 08 95        	ret

However the generated object file does contain the avr2 family information.

ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              REL (Relocatable file)
  Machine:                           Atmel AVR 8-bit microcontroller
  Version:                           0x1
  Entry point address:               0x0
  Start of program headers:          0 (bytes into file)
  Start of section headers:          284 (bytes into file)
  Flags:                             0x82, EF_AVR_ARCH_AVR2, relaxable
  Size of this header:               52 (bytes)
  Size of program headers:           0 (bytes)
  Number of program headers:         0
  Size of section headers:           40 (bytes)
  Number of section headers:         5
  Section header string table index: 1

So the llvm-objdump should automatically select the avr family according to ELF header, other than rely on user's explicit specification.

[llvmbot]

@llvm/issue-subscribers-tools-llvm-objdump

Author: Ben Shi (benshi001)

Given a simple test program: ```c++ int adiw(int a) { return a + 51; } ```

Build it with clang a.c --target=avr -mmcu=avr2 -c -O3, and then disassemble it with llvm-objdump -d a.o, we see

a.o:	file format elf32-avr

Disassembly of section .text:

00000000 <adiw>:
       0: c3 96 08 95  	<unknown>

We must specify --mcpu=avr2 to llvm-objdump explicitly to make adiw correctly decoded.

a.o:	file format elf32-avr

Disassembly of section .text:

00000000 <adiw>:
       0: c3 96        	adiw	r24, 0x33
       2: 08 95        	ret

However the generated object file does contain the avr2 family information.

ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              REL (Relocatable file)
  Machine:                           Atmel AVR 8-bit microcontroller
  Version:                           0x1
  Entry point address:               0x0
  Start of program headers:          0 (bytes into file)
  Start of section headers:          284 (bytes into file)
  Flags:                             0x82, EF_AVR_ARCH_AVR2, relaxable
  Size of this header:               52 (bytes)
  Size of program headers:           0 (bytes)
  Number of program headers:         0
  Size of section headers:           40 (bytes)
  Number of section headers:         5
  Section header string table index: 1

So the llvm-objdump should automatically select the avr family according to ELF header, other than rely on user's explicit specification.

[MaskRay]

I have implemented the following for AArch64 and PowerPC

• AArch64: https://reviews.llvm.org/D128030 --mattr=+all
• powerpc64: https://reviews.llvm.org/D127824 --mcpu=future

[benshi001]

I have implemented the following for AArch64 and PowerPC

• AArch64: https://reviews.llvm.org/D128030 --mattr=+all
• powerpc64: https://reviews.llvm.org/D127824 --mcpu=future

Thanks. I will study your way. What I am hoping are

1. decode supported instructions for a specific avr family
2. still show <unknown> for unsupported instructions

It seems default to --mattr=+all can not meet my expectation.