Integrated perfetto + roctracer (#5)

- hosttrace library automatically collects and merges timestamps for HIP API calls and kernels with the host-side instrumentation
  - mostly eliminates the need for using external rocprof
- added thread_instruction_count in perfetto output
- increased hosttrace min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- miscellaneous cmake updates

* roctracer support

- fully integrated perfetto + roctracer outputs
- thread_instruction_count in perfetto
- increased min_loop_address_range to 512
- disabled instrumenting functions with dynamic callsites by default
- updated timemory submodule

* hosttrace_launch_compiler

- support for using an alternative compiler as needed via launch compiler
- elfio added as submodule (not currently used)
- miscellaneous cmake updates

* README update + host/device categories + misc

- timemory fix for TIMEMORY_ROCTRACER_ENABLED
- transpose fix

* papi_tuple_t -> papi_tot_ins

- minor fix to Findroctracer.cmake

scripts/hosttrace_launch_compiler

@@ -0,0 +1,75 @@
+#!/bin/bash -e
+#
+#   This script allows CMAKE_CXX_COMPILER to be a standard
+#   C++ compiler and hosttrace sets RULE_LAUNCH_COMPILE and
+#   RULE_LAUNCH_LINK in CMake so that all compiler and link
+#   commands are prefixed with this script followed by the
+#   C++ compiler. Thus if $1 == $2 then we know the command
+#   was intended for the C++ compiler and we discard both
+#   $1 and $2 and redirect the command to linker.
+#   If $1 != $2 then we know that the command was not intended
+#   for the C++ compiler and we just discard $1 and launch
+#   the original command. Examples of when $2 will not equal
+#   $1 are 'ar', 'cmake', etc. during the linking phase
+#
+
+# emit a message about the underlying command executed
+: ${DEBUG:=0}
+: ${HOSTTRACE_DEBUG_LAUNCH_COMPILER:=${DEBUG}}
+
+debug-message()
+{
+    if [ "${HOSTTRACE_DEBUG_LAUNCH_COMPILER}" -ne 0 ]; then
+        echo -e "##### $(basename ${BASH_SOURCE[0]}) executing: \"$@\"... #####"
+    fi
+}
+
+# if hosttrace compiler is not passed, someone is probably trying to invoke it directly
+if [ -z "${1}" ]; then
+    echo -e "\n${BASH_SOURCE[0]} was invoked without the hosttrace compiler as the first argument."
+    echo "This script is not indended to be directly invoked by any mechanism other"
+    echo -e "than through a RULE_LAUNCH_COMPILE or RULE_LAUNCH_LINK property set in CMake.\n"
+    exit 1
+fi
+
+# if hosttrace compiler is not passed, someone is probably trying to invoke it directly
+if [ -z "${2}" ]; then
+    echo -e "\n${BASH_SOURCE[0]} was invoked without the C++ compiler as the second argument."
+    echo "This script is not indended to be directly invoked by any mechanism other"
+    echo -e "than through a RULE_LAUNCH_COMPILE or RULE_LAUNCH_LINK property set in CMake.\n"
+    exit 1
+fi
+
+# if there aren't two args, this isn't necessarily invalid, just a bit strange
+if [ -z "${3}" ]; then exit 0; fi
+
+# store the hosttrace compiler
+HOSTTRACE_COMPILER=${1}
+
+# remove the hosttrace compiler from the arguments
+shift
+
+# store the expected C++ compiler
+CXX_COMPILER=${1}
+
+# remove the expected C++ compiler from the arguments
+shift
+
+if [[ "${CXX_COMPILER}" != "${1}" ]]; then
+    debug-message $@
+    # the command does not depend on hosttrace so just execute the command w/o re-directing to ${HOSTTRACE_COMPILER}
+    eval $@
+else
+    # the executable is the C++ compiler, so we need to re-direct to ${HOSTTRACE_COMPILER}
+    if [ ! -f "${HOSTTRACE_COMPILER}" ]; then
+        echo -e "\nError: the compiler redirect for hosttrace was not found at ${HOSTTRACE_COMPILER}\n"
+        exit 1
+    fi
+
+    # discard the compiler from the command
+    shift
+
+    debug-message ${HOSTTRACE_COMPILER} $@
+    # execute ${HOSTTRACE_COMPILER} (again, usually nvcc_wrapper)
+    ${HOSTTRACE_COMPILER} $@
+fi