Merge remote-tracking branch 'origin/develop' into transformer-opts

src/targets/gpu/include/migraphx/gpu/sync_device.hpp

@@ -15,7 +15,7 @@ namespace gpu {
 struct sync_device
 {
     std::string name() const { return "sync_device"; }
-    void apply(module& p) const;
+    void apply(module& m) const;
 };
 } // namespace gpu
 } // namespace MIGRAPHX_INLINE_NS

src/targets/gpu/include/migraphx/gpu/write_literals.hpp

@@ -14,7 +14,7 @@ struct write_literals
     context* ctx = nullptr;
     std::string name() const { return "gpu::write_literals"; }
 
-    void apply(module& p) const;
+    void apply(module& m) const;
 };
 
 } // namespace gpu

src/targets/gpu/schedule_model.cpp

@@ -77,28 +77,28 @@ MIGRAPHX_REGISTER_OP(wait_event)
 MIGRAPHX_REGISTER_OP(set_stream)
 
 std::size_t schedule_model::concurrency() const { return streams; }
-void schedule_model::sched(module& p, instruction_ref ins, std::size_t n) const
+void schedule_model::sched(module& m, instruction_ref ins, std::size_t n) const
 {
     auto last_stream = std::find_if(std::make_reverse_iterator(ins),
-                                    std::make_reverse_iterator(p.begin()),
+                                    std::make_reverse_iterator(m.begin()),
                                     [&](auto&& i) { return i.name() == "gpu::set_stream"; });
-    if(last_stream != std::make_reverse_iterator(p.begin()))
+    if(last_stream != std::make_reverse_iterator(m.begin()))
     {
         auto&& op = any_cast<set_stream>(last_stream->get_operator());
         // If the same stream was set earlier then skip
         if(op.stream == n)
             return;
     }
-    p.insert_instruction(ins, set_stream{n});
+    m.insert_instruction(ins, set_stream{n});
 }
 
-void schedule_model::wait(module& p, instruction_ref ins, std::size_t wait_id) const
+void schedule_model::wait(module& m, instruction_ref ins, std::size_t wait_id) const
 {
-    p.insert_instruction(ins, wait_event{wait_id});
+    m.insert_instruction(ins, wait_event{wait_id});
 }
-void schedule_model::record(module& p, instruction_ref ins, std::size_t wait_id) const
+void schedule_model::record(module& m, instruction_ref ins, std::size_t wait_id) const
 {
-    p.insert_instruction(std::next(ins), record_event{wait_id});
+    m.insert_instruction(std::next(ins), record_event{wait_id});
 }
 
 static std::unordered_map<std::string, std::size_t> create_weight_map()

src/targets/gpu/sync_device.cpp

@@ -8,9 +8,9 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace gpu {
 
-void sync_device::apply(module& p) const
+void sync_device::apply(module& m) const
 {
-    auto last = std::prev(p.end());
+    auto last = std::prev(m.end());
     if(last->name() == "@return")
     {
         auto inputs = last->inputs();
@@ -18,10 +18,10 @@ void sync_device::apply(module& p) const
                return (i->name() == "hip::copy_from_gpu");
            }))
         {
-            auto sync_in = p.insert_instruction(last, make_op("hip::sync_stream"), inputs);
+            auto sync_in = m.insert_instruction(last, make_op("hip::sync_stream"), inputs);
             if(not inputs.empty())
             {
-                p.replace_instruction(inputs.front(), sync_in);
+                m.replace_instruction(inputs.front(), sync_in);
             }
         }
     }

src/targets/gpu/write_literals.cpp

@@ -11,25 +11,25 @@ namespace gpu {
 
 MIGRAPHX_DECLARE_ENV_VAR(MIGRAPHX_COPY_LITERALS)
 
-void write_literals::apply(module& p) const
+void write_literals::apply(module& m) const
 {
     assert(ctx != nullptr);
     std::size_t n = 0;
-    for(auto ins : iterator_for(p))
+    for(auto ins : iterator_for(m))
     {
         if(ins->name() == "@literal")
         {
             if(enabled(MIGRAPHX_COPY_LITERALS{}))
             {
                 literal l  = ins->get_literal();
-                auto pre   = p.add_literal(l);
-                auto alloc = p.insert_instruction(std::next(pre), hip_allocate{l.get_shape()});
-                p.replace_instruction(ins, hip_copy_to_gpu{}, pre, alloc);
+                auto pre   = m.add_literal(l);
+                auto alloc = m.insert_instruction(std::next(pre), hip_allocate{l.get_shape()});
+                m.replace_instruction(ins, hip_copy_to_gpu{}, pre, alloc);
             }
             else
             {
-                std::string id = p.name() + ":@literal:" + std::to_string(n);
-                p.replace_instruction(ins, hip_copy_literal{ins->get_literal(), id});
+                std::string id = m.name() + ":@literal:" + std::to_string(n);
+                m.replace_instruction(ins, hip_copy_literal{ins->get_literal(), id});
                 n++;
             }
         }

tools/install_prereqs.sh

@@ -4,12 +4,20 @@
 
 set -e
 
-#install pip3, rocm-cmake, rocblas and miopen
-apt update && apt install -y python3-pip rocm-cmake rocblas miopen-hip openmp-extras
+export LC_ALL=C.UTF-8
+export LANG=C.UTF-8
+
+
+# Need pip3 and Python headers to build dependencies
+apt update && apt install -y python3-pip python3-dev cmake rocm-cmake rocblas miopen-hip openmp-extras
+
+# Needed for cmake to build various pip packages
+pip3 install setuptools wheel
 
 # install rbuild to build dependencies
 pip3 install https://github.com/RadeonOpenCompute/rbuild/archive/master.tar.gz
 
+
 PREFIX=/usr/local
 REQ_FILE_DIR=""
 if [ "$#" -ge 2 ]; then