From 9e1964ddaaa634e352a4e072be25de3adecf2031 Mon Sep 17 00:00:00 2001
From: Laurent Morichetti <laurent.morichetti@amd.com>
Date: Tue, 28 Apr 2020 22:35:46 -0700
Subject: [PATCH] Make the device binary copy optional

Device binaries that are embedded inside the host binary do not
require a copy. Their lifetime is guaranteed to exceed that of the
loaded executable.

Add a 'make_copy' parameter to amd::Program::addDeviceProgram. If
make_copy is false the original image will be used and will not
get freed when the amd::Program is destroyed.

Change-Id: I7973bb0243f5a2d1b639b8a88445cfe6af919dd7
---
 rocclr/platform/program.cpp | 38 +++++++++++++++++++++----------------
 rocclr/platform/program.hpp |  5 +++--
 2 files changed, 25 insertions(+), 18 deletions(-)

diff --git a/rocclr/platform/program.cpp b/rocclr/platform/program.cpp
index fe9940796b..c961cba550 100644
--- a/rocclr/platform/program.cpp
+++ b/rocclr/platform/program.cpp
@@ -62,8 +62,8 @@ Program::~Program() {
 
   for (const auto& it : binary_) {
     const binary_t& Bin = it.second;
-    if (Bin.first) {
-      delete[] Bin.first;
+    if (std::get<2>(Bin)) {
+      delete[] std::get<0>(Bin);
     }
   }
 
@@ -82,7 +82,7 @@ const Symbol* Program::findSymbol(const char* kernelName) const {
 }
 
 int32_t Program::addDeviceProgram(Device& device, const void* image, size_t length,
-                                 amd::option::Options* options) {
+                                 bool make_copy, amd::option::Options* options) {
   if (image != NULL &&  !amd::isElfMagic((const char*)image)) {
     if (device.settings().useLightning_) {
       return CL_INVALID_BINARY;
@@ -148,22 +148,28 @@ int32_t Program::addDeviceProgram(Device& device, const void* image, size_t leng
   }
 
   if (image != NULL) {
-    uint8_t* memory = binary(rootDev).first;
+    const uint8_t* memory = std::get<0>(binary(rootDev));
     // clone 'binary' (it is owned by the host thread).
     if (memory == NULL) {
-      memory = new (std::nothrow) uint8_t[length];
-      if (memory == NULL) {
-        delete program;
-        return CL_OUT_OF_HOST_MEMORY;
+      if (make_copy) {
+        auto *image_copy = new (std::nothrow) uint8_t[length];
+        if (image_copy == NULL) {
+          delete program;
+          return CL_OUT_OF_HOST_MEMORY;
+        }
+
+        ::memcpy(image_copy, image, length);
+        memory = image_copy;
+      }
+      else {
+        memory = static_cast<const uint8_t*>(image);
       }
 
-      ::memcpy(memory, image, length);
-
       // Save the original image
-      binary_[&rootDev] = std::make_pair(memory, length);
+      binary_[&rootDev] = std::make_tuple(memory, length, make_copy);
     }
 
-    if (!program->setBinary(reinterpret_cast<char*>(memory), length)) {
+    if (!program->setBinary(reinterpret_cast<const char*>(memory), length)) {
       delete program;
       return CL_INVALID_BINARY;
     }
@@ -230,7 +236,7 @@ int32_t Program::compile(const std::vector<Device*>& devices, size_t numHeaders,
     device::Program* devProgram = getDeviceProgram(*it);
     if (devProgram == NULL) {
       const binary_t& bin = binary(*it);
-      retval = addDeviceProgram(*it, bin.first, bin.second, &parsedOptions);
+      retval = addDeviceProgram(*it, std::get<0>(bin), std::get<1>(bin), false, &parsedOptions);
       if (retval != CL_SUCCESS) {
         return retval;
       }
@@ -358,7 +364,7 @@ int32_t Program::link(const std::vector<Device*>& devices, size_t numInputs,
     device::Program* devProgram = getDeviceProgram(*it);
     if (devProgram == NULL) {
       const binary_t& bin = binary(*it);
-      retval = addDeviceProgram(*it, bin.first, bin.second, &parsedOptions);
+      retval = addDeviceProgram(*it, std::get<0>(bin), std::get<1>(bin), false, &parsedOptions);
       if (retval != CL_SUCCESS) {
         return retval;
       }
@@ -506,11 +512,11 @@ int32_t Program::build(const std::vector<Device*>& devices, const char* options,
     device::Program* devProgram = getDeviceProgram(*it);
     if (devProgram == NULL) {
       const binary_t& bin = binary(*it);
-      if (sourceCode_.empty() && (bin.first == NULL)) {
+      if (sourceCode_.empty() && (std::get<0>(bin) == NULL)) {
         retval = false;
         continue;
       }
-      retval = addDeviceProgram(*it, bin.first, bin.second, &parsedOptions);
+      retval = addDeviceProgram(*it, std::get<0>(bin), std::get<1>(bin), false, &parsedOptions);
       if (retval != CL_SUCCESS) {
         return retval;
       }
diff --git a/rocclr/platform/program.hpp b/rocclr/platform/program.hpp
index 1435a7849e..ac31936fde 100644
--- a/rocclr/platform/program.hpp
+++ b/rocclr/platform/program.hpp
@@ -37,6 +37,7 @@
 #include <string>
 #include <vector>
 #include <map>
+#include <tuple>
 #include <utility>
 
 namespace amd {
@@ -79,7 +80,7 @@ class Context;
 //! A collection of binaries for devices in the associated context.
 class Program : public RuntimeObject {
  public:
-  typedef std::pair<uint8_t*, size_t> binary_t;
+  typedef std::tuple<const uint8_t* /*image*/, size_t /*size*/,  bool /*allocated*/> binary_t;
   typedef std::set<Device const*> devicelist_t;
   typedef std::unordered_map<Device const*, binary_t> devicebinary_t;
   typedef std::unordered_map<Device const*, device::Program*> deviceprograms_t;
@@ -167,7 +168,7 @@ class Program : public RuntimeObject {
 
   //! Add a new device program with or without binary image and options.
   int32_t addDeviceProgram(Device&, const void* image = NULL, size_t len = 0,
-                          amd::option::Options* options = NULL);
+                          bool make_copy = true, amd::option::Options* options = NULL);
 
   //! Find the section for the given device. Return NULL if not found.
   device::Program* getDeviceProgram(const Device& device) const;