Re: [PATCH ARM v2] PR69770 -mlong-calls does not affect calls to __gnu_mcount_nc generated by -pg

[Ramana Radhakrishnan <ramana dot radhakrishnan at foss dot arm dot com>]

On 24/03/16 21:02, Charles Baylis wrote:
> When compiling with -mlong-calls and -pg, calls to the __gnu_mcount_nc
> function are not generated as long calls.
> 
> This is the sequel to this patch
>     https://gcc.gnu.org/ml/gcc-patches/2016-02/msg00881.html
> 
> This patch fixes the following problems with the previous patch.
> . Nested functions now work (thanks to Richard E for spotting this)
> . Thumb-1 now works

> 
> This patch works by adding new patterns (one for ARM/Thumb-2 and one
> for Thumb-1) which are placed in the prologue as a placeholder for
> some RTL which describes the address. This is either a SYMBOL_REF for
> targets with MOVW/MOVT, or a literal pool reference for other targets.
> The implementation of ARM_FUNCTION_PROFILER is changed to search for
> this insn so that the the address of the __gnu_mcount_nc function can
> be loaded using an appropriate sequence for the target.

I'm reasonably happy with the approach but there are nits.

> 
> I also tried generating the profiling call sequence directly in the
> prologue, but this requires some unpleasant hacks to prevent spurious
> register pushes from ASM_OUTPUT_REG_PUSH.
> 
> Tested with no new regressions on arm-unknown-linux-gnueabihf on QEMU.
> The generated code sequences have been inspected for normal and nested
> functions on ARM v6, ARM v7, Thumb-1, and Thumb-2 targets.
> 
> This does not fix a regression, so I don't expect to apply it for
> GCC6, is it OK for when stage 1 re-opens.


I'm not sure how much testing coverage you get by just running the testsuite. Doing a profiled bootstrap with -mlong-calls and a regression test run for arm and / or thumb2 would be a useful test to do additionally - Given that this originally came from kernel builds with allyesconfig how important is this to be fixed for GCC 6 ? I'd rather take the fix into GCC 6 to get the kernel building again but that's something we can discuss with the RM's once the issues with the patch are fixed.
> 
> gcc/ChangeLog:
> 
> 2016-03-24  Charles Baylis  <charles.baylis@linaro.org>
> 
>         * config/arm/arm-protos.h (arm_emit_long_call_profile): New function.
>         * config/arm/arm.c (arm_emit_long_call_profile_insn): New function.
>         (arm_expand_prologue): Likewise.
>         (thumb1_expand_prologue): Likewise.
>         (arm_output_long_call_to_profile_func): Likewise.
>         (arm_emit_long_call_profile): Likewise.
>         * config/arm/arm.h: (ASM_OUTPUT_REG_PUSH) Update comment.
>         * config/arm/arm.md (arm_long_call_profile): New pattern.
>         * config/arm/bpabi.h (ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS): New
>         define.
>         * config/arm/thumb1.md (thumb1_long_call_profile): New pattern.
>         * config/arm/unspecs.md (unspecv): Add VUNSPEC_LONG_CALL_PROFILE.
> 
> gcc/testsuite/ChangeLog:
> 
> 2016-03-24  Charles Baylis  <charles.baylis@linaro.org>
> 
>         * gcc.target/arm/pr69770.c: New test.
> 
> 
> 0001-PR69770-mlong-calls-does-not-affect-calls-to-__gnu_m.patch
> 
> 
> From 5a39451f34be9b6ca98b3460bf40d879d6ee61a5 Mon Sep 17 00:00:00 2001
> From: Charles Baylis <charles.baylis@linaro.org>
> Date: Thu, 24 Mar 2016 20:43:25 +0000
> Subject: [PATCH] PR69770 -mlong-calls does not affect calls to __gnu_mcount_nc
>  generated by -pg
> 
> gcc/ChangeLog:
> 
> 2016-03-24  Charles Baylis  <charles.baylis@linaro.org>
> 
>         * config/arm/arm-protos.h (arm_emit_long_call_profile): New function.
>         * config/arm/arm.c (arm_emit_long_call_profile_insn): New function.
>         (arm_expand_prologue): Likewise.
>         (thumb1_expand_prologue): Likewise.
>         (arm_output_long_call_to_profile_func): Likewise.
>         (arm_emit_long_call_profile): Likewise.
>         * config/arm/arm.h: (ASM_OUTPUT_REG_PUSH) Update comment.
>         * config/arm/arm.md (arm_long_call_profile): New pattern.
>         * config/arm/bpabi.h (ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS): New
> 	define.
>         * config/arm/thumb1.md (thumb1_long_call_profile): New pattern.
>         * config/arm/unspecs.md (unspecv): Add VUNSPEC_LONG_CALL_PROFILE.
> 
> gcc/testsuite/ChangeLog:
> 
> 2016-03-24  Charles Baylis  <charles.baylis@linaro.org>
> 
>         * gcc.target/arm/pr69770.c: New test.
> 
> Change-Id: I9b8de01fea083f17f729c3801f83174bedb3b0c6
> 
> diff --git a/gcc/config/arm/arm-protos.h b/gcc/config/arm/arm-protos.h
> index 0083673..324c9f4 100644
> --- a/gcc/config/arm/arm-protos.h
> +++ b/gcc/config/arm/arm-protos.h
> @@ -343,6 +343,7 @@ extern void arm_register_target_pragmas (void);
>  extern void arm_cpu_cpp_builtins (struct cpp_reader *);
>  
>  extern bool arm_is_constant_pool_ref (rtx);
> +void arm_emit_long_call_profile ();
>  
>  /* Flags used to identify the presence of processor capabilities.  */
>  
> diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
> index c868490..040b255 100644
> --- a/gcc/config/arm/arm.c
> +++ b/gcc/config/arm/arm.c
> @@ -21426,6 +21426,22 @@ output_probe_stack_range (rtx reg1, rtx reg2)
>    return "";
>  }
>  
> +static void
> +arm_emit_long_call_profile_insn ()

s/()/(void)

> +{
> +  rtx sym_ref = gen_rtx_SYMBOL_REF (Pmode, "__gnu_mcount_nc");
> +  /* if movt/movw are not available, use a constant pool */
> +  if (!arm_arch_thumb2)

Please don't do that - use TARGET_USE_MOVT && !target_word_relocations please. And better still look at adding !target_word_relocations to the define of TARGET_USE_MOVT.

> +  {
> +    sym_ref = force_const_mem(Pmode, sym_ref);

Space between force_const_mem and `('

> +  }
> +  rtvec vec = gen_rtvec (1, sym_ref);
> +  rtx tmp =
> +    gen_rtx_UNSPEC_VOLATILE (VOIDmode, vec, VUNSPEC_LONG_CALL_PROFILE);
> +  emit_insn (tmp);
> +}
> +

> +
>  /* Generate the prologue instructions for entry into an ARM or Thumb-2
>     function.  */
>  void
> @@ -21789,6 +21805,10 @@ arm_expand_prologue (void)
>        arm_load_pic_register (mask);
>      }
>  
> +  if (crtl->profile && TARGET_LONG_CALLS
> +      && ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS)
> +    arm_emit_long_call_profile_insn ();
> +
>    /* If we are profiling, make sure no instructions are scheduled before
>       the call to mcount.  Similarly if the user has requested no
>       scheduling in the prolog.  Similarly if we want non-call exceptions
> @@ -24985,6 +25005,10 @@ thumb1_expand_prologue (void)
>    if (frame_pointer_needed)
>      thumb_set_frame_pointer (offsets);
>  
> +  if (crtl->profile && TARGET_LONG_CALLS
> +      && ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS)
> +    arm_emit_long_call_profile_insn ();
> +
>    /* If we are profiling, make sure no instructions are scheduled before
>       the call to mcount.  Similarly if the user has requested no
>       scheduling in the prolog.  Similarly if we want non-call exceptions
> @@ -30289,4 +30313,70 @@ arm_sched_fusion_priority (rtx_insn *insn, int max_pri,
>    return;
>  }
>  
> +static void
> +arm_output_long_call_to_profile_func (rtx * operands, bool push_scratch)
> +{

Missing comment above function.

> +  /* operands[0] is the address of the __gnu_mcount_nc function
> +     operands[1] is the scratch register we use to load that address */
> +  if (push_scratch)
> +    output_asm_insn ("push\t{%1}", operands);
> +  output_asm_insn ("push\t{lr}", operands);
> +  if (GET_CODE (operands[0]) == SYMBOL_REF)
> +    {
> +      output_asm_insn ("movw\t%1, #:lower16:%c0", operands);
> +      output_asm_insn ("movt\t%1, #:upper16:%c0", operands);
> +    }
> +  else
> +    {
> +      output_asm_insn ("ldr\t%1, %0", operands);
> +    }
> +  if (!arm_arch5)
> +    {
> +      output_asm_insn ("mov\tlr, pc", operands);
> +      output_asm_insn ("mov\tpc, %1", operands);
> +    }
> +  else
> +    output_asm_insn ("blx\t%1", operands);
> +  if (push_scratch)
> +    output_asm_insn ("pop\t{%1}", operands);
> +}
> +


> +void
> +arm_emit_long_call_profile()

s/()/ (void)

Missing comment.

> +{
> +  rtx alcp = NULL;
> +  rtx operands[2];
> +  bool push_scratch;
> +  /* find the arm_long_call_profile */
> +  for (rtx_insn * insn = get_insns (); insn; insn = NEXT_INSN (insn))
> +    {
> +      if (NOTE_P (insn) && NOTE_KIND (insn) == NOTE_INSN_PROLOGUE_END)
> +	break;
> +      if (INSN_CODE (insn) == CODE_FOR_arm_long_call_profile
> +	  || INSN_CODE (insn) == CODE_FOR_thumb1_long_call_profile)
> +	{
> +	  alcp = PATTERN (insn);
> +	  break;
> +	}
> +    }
> +  gcc_assert (alcp);
> +
> +  operands[0] = XEXP (XEXP (alcp, 0), 0);
> +  if (TARGET_32BIT)
> +    {
> +      operands[1] = gen_rtx_REG (SImode, IP_REGNUM);
> +      push_scratch = false;
> +    }
> +  else
> +    {
> +      /* for nested functions, we can set push_scratch to false, since
> +	 final.c:profile_function.c and ASM_OUTPUT_REG_PUSH preserve it as
> +         part of the sequence to preserve ip across the call to the 
> +         profiling function. */
> +      operands[1] = gen_rtx_REG (SImode, R0_REGNUM + 7);
> +      push_scratch = !IS_NESTED (arm_current_func_type ());
> +    }
> +  arm_output_long_call_to_profile_func (operands, push_scratch);
> +}
> +
>  #include "gt-arm.h"
> diff --git a/gcc/config/arm/arm.h b/gcc/config/arm/arm.h
> index 6352140..44f1c64 100644
> --- a/gcc/config/arm/arm.h
> +++ b/gcc/config/arm/arm.h
> @@ -2044,7 +2044,9 @@ extern int making_const_table;
>     that ASM_OUTPUT_REG_PUSH will be matched with ASM_OUTPUT_REG_POP, and
>     that r7 isn't used by the function profiler, so we can use it as a
>     scratch reg.  WARNING: This isn't safe in the general case!  It may be
> -   sensitive to future changes in final.c:profile_function.  */
> +   sensitive to future changes in final.c:profile_function. This is also
> +   relied on in arm_emit_long_call_profile() which assumes r7 can be
> +   used as a scratch register to load the address of the function profiler.  */
>  #define ASM_OUTPUT_REG_PUSH(STREAM, REGNO)		\
>    do							\
>      {							\
> diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
> index 47171b9..beb44d5 100644
> --- a/gcc/config/arm/arm.md
> +++ b/gcc/config/arm/arm.md
> @@ -11424,6 +11424,15 @@
>    DONE;
>  })
>  
> +(define_insn "arm_long_call_profile"
> +  [(unspec_volatile [(match_operand:SI 0 "general_operand" "ji,m")
> +                    ] VUNSPEC_LONG_CALL_PROFILE)]
> +  "TARGET_32BIT"
> +  "%@ arm_long_call_profile"
> +  [(set_attr "arm_pool_range" "*,4096")
> +   (set_attr "arm_neg_pool_range" "*,4084")]
> +)

Please set the length of this insn appropriately. IIUC this is the length of the instructions from FUNCTION_PROFILER, though I suspect we could push all of this out into the RTL stream. Underestimating the length of this insn can cause issues with the minipool placement code later especially with -pg.

> +
>  ;; Vector bits common to IWMMXT and Neon
>  (include "vec-common.md")
>  ;; Load the Intel Wireless Multimedia Extension patterns
> diff --git a/gcc/config/arm/bpabi.h b/gcc/config/arm/bpabi.h
> index 5d6c4ed..7acf66b 100644
> --- a/gcc/config/arm/bpabi.h
> +++ b/gcc/config/arm/bpabi.h
> @@ -174,11 +174,18 @@
>  
>  #undef  NO_PROFILE_COUNTERS
>  #define NO_PROFILE_COUNTERS 1
> +#undef  ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS
> +#define ARM_FUNCTION_PROFILER_SUPPORTS_LONG_CALLS 1

Why do we need this ?

>  #undef  ARM_FUNCTION_PROFILER
>  #define ARM_FUNCTION_PROFILER(STREAM, LABELNO)  			\
>  {									\
> -  fprintf (STREAM, "\tpush\t{lr}\n");					\
> -  fprintf (STREAM, "\tbl\t__gnu_mcount_nc\n");				\
> +  if (TARGET_LONG_CALLS)						\
> +  { 									\
> +    arm_emit_long_call_profile();					\
> +  } else {								\
> +    fprintf (STREAM, "\tpush\t{lr}\n");					\
> +    fprintf (STREAM, "\tbl\t__gnu_mcount_nc\n");			\
> +  }									\
>  }
>  

While you are here delete the definition of FUNCTION_PROFILER that is conditional on THUMB_FUNCTION_PROFILER, looks like that code is dead.

>  #undef SUBTARGET_FRAME_POINTER_REQUIRED
> diff --git a/gcc/config/arm/thumb1.md b/gcc/config/arm/thumb1.md
> index 072ed4d..9b1de5f 100644
> --- a/gcc/config/arm/thumb1.md
> +++ b/gcc/config/arm/thumb1.md
> @@ -1798,7 +1798,7 @@
>    [(unspec_volatile [(match_operand:SI 0 "s_register_operand" "l")]
>  		    VUNSPEC_EH_RETURN)
>     (clobber (match_scratch:SI 1 "=&l"))]
> -  "TARGET_THUMB1"
> +  "TARGET_THUMB1 && 0"

This means it wasn't tested for Thumb1 or you had to turn this off to get proper results for Thumb1 ?

Please retest.

>    "#"
>    "&& reload_completed"
>    [(const_int 0)]
> @@ -1809,4 +1809,13 @@
>    }"
>    [(set_attr "type" "mov_reg")]
>  )
> +
> +(define_insn "thumb1_long_call_profile"
> +  [(unspec_volatile [(match_operand:SI 0 "general_operand" "j,m")
> +                    ] VUNSPEC_LONG_CALL_PROFILE)]
> +  "TARGET_THUMB1"
> +  "%@ thumb1_long_call_profile"
> +  [(set_attr "pool_range" "1018")]
> +)
> +
>  

Please set length accordingly as well as check on the pool_range here.

> diff --git a/gcc/config/arm/unspecs.md b/gcc/config/arm/unspecs.md
> index 5744c62..d7ddc3a 100644
> --- a/gcc/config/arm/unspecs.md
> +++ b/gcc/config/arm/unspecs.md
> @@ -148,6 +148,7 @@
>    VUNSPEC_GET_FPSCR	; Represent fetch of FPSCR content.
>    VUNSPEC_SET_FPSCR	; Represent assign of FPSCR content.
>    VUNSPEC_PROBE_STACK_RANGE ; Represent stack range probing.
> +  VUNSPEC_LONG_CALL_PROFILE ; Represent a long call to profile function
>  ])
>  
>  ;; Enumerators for NEON unspecs.
> diff --git a/gcc/testsuite/gcc.target/arm/pr69770.c b/gcc/testsuite/gcc.target/arm/pr69770.c
> new file mode 100644
> index 0000000..61e5c6d
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/pr69770.c
> @@ -0,0 +1,9 @@
> +/* { dg-do compile } */
> +/* { dg-options "-pg -mlong-calls" } */
> +
> +extern void g(void);
> +
> +int f() { g(); return 0; }
> +
> +/* { dg-final { scan-assembler-not "bl\[ \t\]+__gnu_mcount_nc" } } */
> +/* { dg-final { scan-assembler "__gnu_mcount_nc" } } */
> -- 1.9.1
>