From 00efdc1cd6f3140efcefb1b6f9de86371d15dbba Mon Sep 17 00:00:00 2001
From: Jeremy Newton <Jeremy.Newton@amd.com>
Date: Wed, 6 Apr 2022 11:07:35 -0400
Subject: [PATCH] SWDEV-323669 - Improve arch detection

- Clean up detection by using visual studio macros to detect arch; I
  didn't list all possible ARM platforms (can be done later if desired)
- Fixed two incorrect uses of !defined(ATI_ARCH_ARM) to instead use
  defined(ATI_ARCH_X86), as they contain X86 specific code
- Fixed one use of __ARM_ARCH_7A__ to use ATI_ARCH_ARM instead

This is an improvement to the fixes in the last patch for SWDEV-323669

Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Change-Id: I8568167293c34ad5331902105877f3ab6e25acb3
---
 rocclr/include/top.hpp | 12 ++++--------
 rocclr/os/os.cpp       |  2 +-
 rocclr/os/os_posix.cpp |  6 +++---
 3 files changed, 8 insertions(+), 12 deletions(-)

diff --git a/rocclr/include/top.hpp b/rocclr/include/top.hpp
index 70273510c9..d6aa00c8d3 100644
--- a/rocclr/include/top.hpp
+++ b/rocclr/include/top.hpp
@@ -21,15 +21,11 @@
 #ifndef TOP_HPP_
 #define TOP_HPP_
 
-#if defined(__GNUC__)
-#if defined(__arm__) || defined(__aarch64__)
+#if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__arm) || defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || defined(_M_ARM64)
 #define ATI_ARCH_ARM
-#elif defined(__x86__) || defined(__x86_64__)
+#elif defined(i386) || defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__x86__) || defined(__x86_64) || defined(__x86_64__) || defined(__amd64) || defined(__amd64__) || defined(_M_X64) || defined(_M_AMD64)
 #define ATI_ARCH_X86
 #endif
-#else /*!__GNUC__*/
-#define ATI_ARCH_X86
-#endif /*!__GNUC__*/
 
 #if defined(ATI_ARCH_ARM)
 #define __EXPORTED_HEADERS__ 1
@@ -51,9 +47,9 @@
 #define CL_DEVICE_HOST_UNIFIED_MEMORY 0x1035
 #endif
 
-#if !defined(ATI_ARCH_ARM)
+#if defined(ATI_ARCH_X86)
 #include <xmmintrin.h>
-#endif /*!ATI_ARCH_ARM*/
+#endif /*ATI_ARCH_X86*/
 
 #include <atomic>
 #include <cstdint>
diff --git a/rocclr/os/os.cpp b/rocclr/os/os.cpp
index f5ea547ddf..0437f7f1d6 100644
--- a/rocclr/os/os.cpp
+++ b/rocclr/os/os.cpp
@@ -112,7 +112,7 @@ int Os::processorCount_ = 0;
 void Os::spinPause() {
 #if defined(ATI_ARCH_X86)
   _mm_pause();
-#elif defined(__ARM_ARCH_7A__)
+#elif defined(ATI_ARCH_ARM)
   __asm__ __volatile__("yield");
 #endif
 }
diff --git a/rocclr/os/os_posix.cpp b/rocclr/os/os_posix.cpp
index dd1e61fccf..aa43a7730b 100644
--- a/rocclr/os/os_posix.cpp
+++ b/rocclr/os/os_posix.cpp
@@ -701,14 +701,14 @@ void Os::setCurrentStackPtr(address sp) {
   sp -= sizeof(void*);
   *(void**)sp = __builtin_return_address(0);
 
-#if defined(ATI_ARCH_ARM)
-  assert(!"Unimplemented");
-#else
+#if defined(ATI_ARCH_X86)
   __asm__ __volatile__(
 #if !defined(OMIT_FRAME_POINTER)
       LP64_SWITCH("movl (%%ebp),%%ebp;", "movq (%%rbp),%%rbp;")
 #endif  // !OMIT_FRAME_POINTER
           LP64_SWITCH("movl %0,%%esp; ret;", "movq %0,%%rsp; ret;")::"r"(sp));
+#else
+  assert(!"Unimplemented");
 #endif
 }