E2K: added initial support for MCST Elbrus 2000

src/nvcore/Debug.cpp

@@ -668,6 +668,13 @@ namespace
 #    elif NV_CPU_AARCH64
         ucontext_t * ucp = (ucontext_t *)secret;
         return (void *) ucp->uc_mcontext.pc;
+#   elif NV_CPU_E2K /* MCST Elbrus 2000 */
+        // e2k has 3 stacks - Procedure Stack (PS), Procedure Chain Stack (PCS) and User Stack (US)
+        // CR0 and CR1 (Chain Register) are the 128-bit registers of the Procedure Chain Stack (PCS)
+        // CR's divided into _HI and _LO 64-bit parts (as in x86, for example, AX is divided into AH and AL)
+        // CR0_HI stores an Instruction Pointer
+        ucontext_t * ucp = (ucontext_t *)secret;
+        return (void *) ucp->uc_mcontext.cr0_hi;
 #    else
 #      error "Unknown CPU"
 #    endif

src/nvcore/nvcore.h

@@ -98,6 +98,7 @@
 // NV_CPU_PPC
 // NV_CPU_ARM
 // NV_CPU_ARM_64
+// NV_CPU_E2K
 
 #define NV_CPU_STRING   POSH_CPU_STRING
 
@@ -112,6 +113,8 @@
 #   define NV_CPU_ARM 1
 #elif defined POSH_CPU_AARCH64
 #   define NV_CPU_ARM_64 1
+#elif defined POSH_CPU_E2K
+#   define NV_CPU_E2K 1
 #else
 #   error "Unsupported CPU"
 #endif

src/nvthread/Atomic.h

@@ -61,7 +61,7 @@ namespace nv {
 #elif POSH_CPU_STRONGARM || POSH_CPU_AARCH64
         // need more specific cpu type for armv7?
         // also utilizes a full barrier
-        // currently treating laod like x86 - this could be wrong
+        // currently treating load like x86 - this could be wrong
         
         // this is the easiest but slowest way to do this
         nvCompilerReadWriteBarrier();
@@ -73,6 +73,16 @@ namespace nv {
         // also utilizes a full barrier
         // currently treating load like x86 - this could be wrong
 
+        // this is the easiest but slowest way to do this
+        nvCompilerReadWriteBarrier();
+		uint32 ret = *ptr; // replace with ldrex?
+        nvCompilerReadWriteBarrier();
+        return ret;
+#elif POSH_CPU_E2K
+        // need more specific cpu type for e2k?
+        // also utilizes a full barrier
+        // currently treating load like x86 - this could be wrong
+
         // this is the easiest but slowest way to do this
         nvCompilerReadWriteBarrier();
         uint32 ret = *ptr; // replace with ldrex?
@@ -102,6 +112,11 @@ namespace nv {
         nvCompilerReadWriteBarrier();
 		*ptr = value; //strex?
 		nvCompilerReadWriteBarrier();
+#elif POSH_CPU_E2K
+        // this is the easiest but slowest way to do this
+        nvCompilerReadWriteBarrier();
+        *ptr = value; //strex?
+        nvCompilerReadWriteBarrier();
 #else
 #error "Atomics not implemented."
 #endif