From cc602061ee860b041013397e27a036b85cd87b09 Mon Sep 17 00:00:00 2001
From: Geza Lore <gezalore@gmail.com>
Date: Mon, 18 Jan 2016 00:21:51 +0100
Subject: [PATCH] x86inc: Add debug symbols indicating sizes of compiled
 functions

Some debuggers/profilers use this metadata to determine which function a
given instruction is in; without it they get can confused by local labels
(if you haven't stripped those). On the other hand, some tools are still
confused even with this metadata. e.g. this fixes `gdb`, but not `perf`.

Currently only implemented for ELF.

Signed-off-by: Anton Khirnov <anton@khirnov.net>
---
 libavcodec/x86/proresdsp.asm    |  2 +-
 libavutil/x86/x86inc.asm        | 23 +++++++++++++++++++++++
 tests/checkasm/x86/checkasm.asm |  8 ++++----
 3 files changed, 28 insertions(+), 5 deletions(-)

diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/x86/proresdsp.asm
index a0e97b3951..5a329cb898 100644
--- a/libavcodec/x86/proresdsp.asm
+++ b/libavcodec/x86/proresdsp.asm
@@ -54,7 +54,7 @@ cextern pw_8
 cextern pw_512
 cextern pw_1019
 
-section .text align=16
+SECTION .text
 
 ; interleave data while maintaining source
 ; %1=type, %2=dstlo, %3=dsthi, %4=src, %5=interleave
diff --git a/libavutil/x86/x86inc.asm b/libavutil/x86/x86inc.asm
index 8a74830a20..20ef7b8a19 100644
--- a/libavutil/x86/x86inc.asm
+++ b/libavutil/x86/x86inc.asm
@@ -634,6 +634,7 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
     %else
         rep ret
     %endif
+    annotate_function_size
 %endmacro
 
 %define last_branch_adr $$
@@ -642,6 +643,7 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
         times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ == last_branch_adr.
     %endif
     ret
+    annotate_function_size
 %endmacro
 
 %macro BRANCH_INSTR 0-*
@@ -666,6 +668,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
     %elif %2
         jmp %1
     %endif
+    annotate_function_size
 %endmacro
 
 ;=============================================================================
@@ -687,6 +690,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
     cglobal_internal 0, %1 %+ SUFFIX, %2
 %endmacro
 %macro cglobal_internal 2-3+
+    annotate_function_size
     %if %1
         %xdefine %%FUNCTION_PREFIX private_prefix
         %xdefine %%VISIBILITY hidden
@@ -700,6 +704,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
         CAT_XDEFINE cglobaled_, %2, 1
     %endif
     %xdefine current_function %2
+    %xdefine current_function_section __SECT__
     %if FORMAT_ELF
         global %2:function %%VISIBILITY
     %else
@@ -748,6 +753,24 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae,
     [SECTION .note.GNU-stack noalloc noexec nowrite progbits]
 %endif
 
+; Tell debuggers how large the function was.
+; This may be invoked multiple times per function; we rely on later instances overriding earlier ones.
+; This is invoked by RET and similar macros, and also cglobal does it for the previous function,
+; but if the last function in a source file doesn't use any of the standard macros for its epilogue,
+; then its size might be unspecified.
+%macro annotate_function_size 0
+    %ifdef __YASM_VER__
+        %ifdef current_function
+            %if FORMAT_ELF
+                current_function_section
+                %%ecf equ $
+                size current_function %%ecf - current_function
+                __SECT__
+            %endif
+        %endif
+    %endif
+%endmacro
+
 ; cpuflags
 
 %assign cpuflags_mmx      (1<<0)
diff --git a/tests/checkasm/x86/checkasm.asm b/tests/checkasm/x86/checkasm.asm
index 52d10aec5c..55212fc24b 100644
--- a/tests/checkasm/x86/checkasm.asm
+++ b/tests/checkasm/x86/checkasm.asm
@@ -66,14 +66,14 @@ cextern fail_func
 ;-----------------------------------------------------------------------------
 cglobal stack_clobber, 1,2
     ; Clobber the stack with junk below the stack pointer
-    %define size (max_args+6)*8
-    SUB  rsp, size
-    mov   r1, size-8
+    %define argsize (max_args+6)*8
+    SUB  rsp, argsize
+    mov   r1, argsize-8
 .loop:
     mov [rsp+r1], r0
     sub   r1, 8
     jge .loop
-    ADD  rsp, size
+    ADD  rsp, argsize
     RET
 
 %if WIN64