Add clang to compiler.h

[Ryzee119]

Describe the PR
I'm using the clang compiler which for our purposes the compiler attributes match GCC from what I can read.

Although TU_ATTR_WEAK had to be changed to __attribute__ ((weak_import)) or I was getting lld: error: duplicate symbol errors on all the TU_ATTR_WEAK functions.

[hathach]

can you do the #ifdef #else instead of undef (ust move this up). undef is rather confusing since people may only look at the first define.

[Ryzee119]

Just to clarify @hathach

Like this?

#if defined(__GNUC__) || defined (__clang__)
  #define TU_ATTR_ALIGNED(Bytes)        __attribute__ ((aligned(Bytes)))
  #define TU_ATTR_SECTION(sec_name)     __attribute__ ((section(#sec_name)))
  #define TU_ATTR_PACKED                __attribute__ ((packed))
  #if defined(__GNUC__)
    #define TU_ATTR_WEAK                  __attribute__ ((weak))
  #else
    #define TU_ATTR_WEAK                  __attribute__ ((weak_import))
  #endif
... 
...

[hathach]

yeah, but the otherway around

  #if defined(__clang__)
    #define TU_ATTR_WEAK                  __attribute__ ((weak_import))
  #else
    #define TU_ATTR_WEAK                  __attribute__ ((weak))
  #endif

[Ryzee119]

All fixed :)
Not sure why pre-commit CI is failing 🤔

[hathach]

unit test probably compile using clang, and has affected by this attribute. I will check later.

[hathach]

still haven't got time to test this out, will be off for a couple of week for TET holidays. Please be patient, thank you.

[hathach]

@Ryzee119 I am back from TET holidays, which compiler you are using. If it is arm clang, I could give it a try (was planning to support it anwyay).

[hathach]

recently I switch to use weak as the default implementation of callback to be compatible with keil #2239

https://github.com/hathach/tinyusb/blob/master/src/device/usbd.c#L48

the other callback header can work with weak_import, but this style seems to require actual weak and cause the compile issue with unit-test (since test include mocking of dcd layer).

Maybe I should use weakref/alias or unify them all to default implemenation, will revise later. We probably need to keep this open until then. Thank you.

[HiFiPhile]

How it goes with #2606 ?

[hathach]

so far I only tested arm-clang with cdc_msc example, it doesn't seem to need weak_import yet, but I admit we need more testinng. the conlfict with unit-test is resolved (unit test switch back to gcc)

[Ryzee119]

So sorry i missed this discussion somehow!

I'm using clang + llvm cross compiling for x86 type embedded system with OHCI backend.

recently I switch to use weak as the default implementation of callback to be compatible with keil #2239

^ Yep this does the trick for me. I was having problems with the OHCI tusb_app_virt_to_phys/tusb_app_phys_to_virt weak functions.

So I just need this patch and implement weak stub functions.

+++ b/src/common/tusb_compiler.h
@@ -123,7 +123,7 @@
 //--------------------------------------------------------------------+

 // TODO refactor since __attribute__ is supported across many compiler
-#if defined(__GNUC__)
+#if defined(__GNUC__) || defined (__clang__)

diff --git a/src/portable/ohci/ohci.c b/src/portable/ohci/ohci.c
index c59d475..715ac63 100644
--- a/src/portable/ohci/ohci.c
+++ b/src/portable/ohci/ohci.c

 TU_ATTR_ALWAYS_INLINE static inline void *_phys_addr(void *virtual_address)
 {
-  if (tusb_app_virt_to_phys) return tusb_app_virt_to_phys(virtual_address);
-  return virtual_address;
+  return tusb_app_virt_to_phys(virtual_address);
 }

 TU_ATTR_ALWAYS_INLINE static inline void *_virt_addr(void *physical_address)
 {
-  if (tusb_app_phys_to_virt) return tusb_app_phys_to_virt(physical_address);
-  return physical_address;
+  return tusb_app_phys_to_virt(physical_address);
 }

Ill probably supersede this PR. I need to stick these two stubs somewhere for my needs. Can you suggest a good location?

TU_ATTR_WEAK void* tusb_app_virt_to_phys(void *virt_addr) {
  return virt_addr;
}

TU_ATTR_WEAK void* tusb_app_phys_to_virt(void *phys_addr) {
  return phys_addr;
}