[OpenMP][NFC] Reorganize the ompx::mapping layer in the GPU runtime

This change makes the naming more consistent, I hope.

openmp/libomptarget/DeviceRTL/include/Debug.h

@@ -31,6 +31,9 @@ void __assert_fail(const char *expr, const char *msg, const char *file,
     else                                                                       \
       __assert_assume(expr);                                                   \
   }
+#define UNREACHABLE(msg)                                                       \
+  PRINT(msg);                                                                  \
+  __builtin_trap();
 
 ///}
 

openmp/libomptarget/DeviceRTL/include/Mapping.h

@@ -18,6 +18,12 @@ namespace ompx {
 
 namespace mapping {
 
+enum {
+  DIM_X = 0,
+  DIM_Y = 1,
+  DIM_Z = 2,
+};
+
 #pragma omp begin declare target device_type(nohost)
 
 inline constexpr uint32_t MaxThreadsPerTeam = 1024;
@@ -63,34 +69,38 @@ LaneMaskTy lanemaskGT();
 /// Return the thread Id in the warp, in [0, getWarpSize()).
 uint32_t getThreadIdInWarp();
 
-/// Return the thread Id in the block, in [0, getBlockSize()).
-uint32_t getThreadIdInBlock();
-
-/// Return the warp id in the block.
-uint32_t getWarpId();
-
 /// Return the warp size, thus number of threads in the warp.
 uint32_t getWarpSize();
 
+/// Return the warp id in the block, in [0, getNumberOfWarpsInBlock()]
+uint32_t getWarpIdInBlock();
+
 /// Return the number of warps in the block.
 uint32_t getNumberOfWarpsInBlock();
 
-/// Return the block Id in the kernel, in [0, getKernelSize()).
-uint32_t getBlockId();
+/// Return the thread Id in the block, in [0, getNumberOfThreadsInBlock(Dim)).
+uint32_t getThreadIdInBlock(int32_t Dim = DIM_X);
 
 /// Return the block size, thus number of threads in the block.
-///
-/// Note: The version taking \p IsSPMD mode explicitly can be used during the
-/// initialization of the target region, that is before `mapping::isSPMDMode()`
-/// can be called by any thread other than the main one.
-uint32_t getBlockSize();
-uint32_t getBlockSize(bool IsSPMD);
+uint32_t getNumberOfThreadsInBlock(int32_t Dim = DIM_X);
+
+/// Return the block Id in the kernel, in [0, getNumberOfBlocksInKernel(Dim)).
+uint32_t getBlockIdInKernel(int32_t Dim = DIM_X);
 
 /// Return the number of blocks in the kernel.
-uint32_t getNumberOfBlocks();
+uint32_t getNumberOfBlocksInKernel(int32_t Dim = DIM_X);
 
 /// Return the kernel size, thus number of threads in the kernel.
-uint32_t getKernelSize();
+uint32_t getNumberOfThreadsInKernel();
+
+/// Return the maximal number of threads in the block usable for a team (=
+/// parallel region).
+///
+/// Note: The version taking \p IsSPMD mode explicitly can be used during the
+/// initialization of the target region, that is before `mapping::isSPMDMode()`
+/// can be called by any thread other than the main one.
+uint32_t getMaxTeamThreads();
+uint32_t getMaxTeamThreads(bool IsSPMD);
 
 /// Return the number of processing elements on the device.
 uint32_t getNumberOfProcessorElements();

openmp/libomptarget/DeviceRTL/src/Debug.cpp

@@ -40,7 +40,8 @@ void __assert_fail(const char *expr, const char *msg, const char *file,
 DebugEntryRAII::DebugEntryRAII(const char *File, const unsigned Line,
                                const char *Function) {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
-      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0) {
+      mapping::getThreadIdInBlock() == 0 &&
+      mapping::getBlockIdInKernel() == 0) {
 
     uint16_t &Level =
         state::getKernelEnvironment().DynamicEnv->DebugIndentionLevel;
@@ -56,7 +57,8 @@ DebugEntryRAII::DebugEntryRAII(const char *File, const unsigned Line,
 
 DebugEntryRAII::~DebugEntryRAII() {
   if (config::isDebugMode(config::DebugKind::FunctionTracing) &&
-      mapping::getThreadIdInBlock() == 0 && mapping::getBlockId() == 0) {
+      mapping::getThreadIdInBlock() == 0 &&
+      mapping::getBlockIdInKernel() == 0) {
     uint16_t &Level =
         state::getKernelEnvironment().DynamicEnv->DebugIndentionLevel;
     Level--;

openmp/libomptarget/DeviceRTL/src/Kernel.cpp

@@ -107,10 +107,11 @@ int32_t __kmpc_target_init(KernelEnvironmentTy &KernelEnvironment) {
   // reaches its corresponding synchronize::threads call: that would permit all
   // active worker threads to proceed before the main thread has actually set
   // state::ParallelRegionFn, and then they would immediately quit without
-  // doing any work.  mapping::getBlockSize() does not include any of the main
-  // thread's warp, so none of its threads can ever be active worker threads.
+  // doing any work.  mapping::getMaxTeamThreads() does not include any of the
+  // main thread's warp, so none of its threads can ever be active worker
+  // threads.
   if (UseGenericStateMachine &&
-      mapping::getThreadIdInBlock() < mapping::getBlockSize(IsSPMD)) {
+      mapping::getThreadIdInBlock() < mapping::getMaxTeamThreads(IsSPMD)) {
     genericStateMachine(KernelEnvironment.Ident);
   } else {
     // Retrieve the work function just to ensure we always call