merge changes from the develop branch

.clang-tidy

@@ -3,6 +3,8 @@ CheckOptions:
     value:           '::std::async;::std::launder;::std::remove;::std::remove_if;::std::unique;::std::unique_ptr::release;::std::basic_string::empty;::std::vector::empty;::std::find;::std::find_if;::std::find_if_not;::std::all_of;::std::any_of;::std::none_of;::std::count;::std::count_if;::std::mismatch;::std::find_end;::std::find_first_of;::std::adjacent_find;::std::search;::std::search_n;::std::nth_element;::std::lower_bound;::std::upper_bound;::std::binary_search;::std::equal_range;::std::max;::std::max_element;::std::min;::std::min_element;::std::minmax;::std::minmax_element;::std::equal;::std::lexicographical_compare;::std::accumulate;::std::inner_product'
   - key:             cppcoreguidelines-macro-usage.AllowedRegexp
     value:           'DEBUG|FALLTHROUGH|_THROW|_REQUIRES|_DECLARE_|_VISIT_|_GENERATE_|_DETAIL_|_MANAGE_PTR|_MATCHER|DEVICE_SHARED'
+  - key:             cppcoreguidelines-narrowing-conversions.WarnOnFloatingPointNarrowingConversion
+    value:           0
   - key:             modernize-loop-convert.MinConfidence
     value:           risky   
   - key:             modernize-loop-convert.NamingStyle

CMakeLists.txt

@@ -56,6 +56,7 @@ rocm_enable_clang_tidy(
         -clang-diagnostic-extern-c-compat
         -clang-diagnostic-disabled-macro-expansion
         -clang-diagnostic-unused-command-line-argument
+        -cppcoreguidelines-explicit-virtual-functions
         -cppcoreguidelines-pro-bounds-array-to-pointer-decay
         -cppcoreguidelines-pro-bounds-constant-array-index
         -cppcoreguidelines-pro-bounds-pointer-arithmetic
@@ -83,6 +84,7 @@ rocm_enable_clang_tidy(
         -modernize-pass-by-value
         -modernize-use-default-member-init
         -modernize-use-transparent-functors
+        -performance-type-promotion-in-math-fn
         -readability-braces-around-statements
         -readability-else-after-return
         -readability-named-parameter

Dockerfile

@@ -44,13 +44,14 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     rm -rf /var/lib/apt/lists/*
 
 # Install cget
-RUN pip install cget
+# RUN pip install cget
+RUN pip install https://github.com/pfultz2/cget/archive/57b3289000fcdb3b7e424c60a35ea09bc44d8538.tar.gz
 
 # Install rclone
 RUN pip install https://github.com/pfultz2/rclone/archive/master.tar.gz
 
 # Install hcc
-RUN rclone -b roc-2.0.x -c 757fb492517b80e7c86338af5fc1a43d63cb25a9 https://github.com/RadeonOpenCompute/hcc.git /hcc
+RUN rclone -b roc-2.3.x -c fd93baed7dcc4fe8019b5fdc90213bfe7c298245 https://github.com/RadeonOpenCompute/hcc.git /hcc
 RUN cget -p $PREFIX install hcc,/hcc
 
 # Use hcc
@@ -73,3 +74,8 @@ ENV LD_LIBRARY_PATH=$PREFIX/lib
 # Install doc requirements
 ADD doc/requirements.txt /doc-requirements.txt
 RUN pip install -r /doc-requirements.txt
+
+# Setup ubsan environment to printstacktrace
+RUN ln -s /usr/bin/llvm-symbolizer-5.0 /usr/local/bin/llvm-symbolizer
+ENV UBSAN_OPTIONS=print_stacktrace=1
+ENV ASAN_OPTIONS=detect_stack_use_after_return=1:check_initialization_order=1:strict_init_order=1

README.md

@@ -12,14 +12,18 @@ AMD's graph optimization engine.
 
 ## Installing the dependencies
 
-The dependencies can be installed with the `install_deps.cmake`, script: `cmake -P install_deps.cmake`.
-
-This will install by default to `/usr/local` but it can be installed in another location with `--prefix` argument:
+Dependencies can be installed using the ROCm build tool [rbuild](https://github.com/RadeonOpenCompute/rbuild).
 
+To install rbuild:
 ```
-cmake -P install_deps.cmake --prefix /some/local/dir
+pip install https://github.com/RadeonOpenCompute/rbuild/archive/master.tar.gz
 ```
 
+To build dependencies along with MIGraphX
+```
+rbuild build -d depend --cxx=/opt/rocm/bin/hcc
+```
+This builds dependencies in the subdirectory named depend and then builds MIGraphX using these dependencies.
 
 ## Building MIGraphX from source
 

cmake/EnableCompilerWarnings.cmake

@@ -90,6 +90,7 @@ else()
                 -Wno-double-promotion
                 -Wno-exit-time-destructors
                 -Wno-extra-semi
+                -Wno-extra-semi-stmt
                 -Wno-float-conversion
                 -Wno-gnu-anonymous-struct
                 -Wno-gnu-zero-variadic-macro-arguments
@@ -104,7 +105,7 @@ else()
         else()
             list(APPEND CMAKE_COMPILER_WARNINGS
                 -Wno-missing-field-initializers
-                -Wno-deprecated-declarations
+                # -Wno-deprecated-declarations
             )
         endif()
         add_definitions(${CMAKE_COMPILER_WARNINGS})

cppcheck.rules

@@ -89,49 +89,76 @@
         <summary>Use manage pointer for resource management</summary>
     </message>
 </rule>
+<rule>
+    <tokenlist>raw</tokenlist>
+    <pattern><![CDATA[hipLaunchKernelGGL \( (?!\( \w+ < \w+ > \))]]></pattern>
+    <message>
+        <id>UseDeviceLaunch</id>
+        <severity>style</severity>
+        <summary>Use device::launch instead</summary>
+    </message>
+</rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern>! !</pattern>
+    <pattern><![CDATA[if (\([^()]*(?-1)*[^()]*\)) { [^{}]* (return|throw|break|continue) [^;]* ; } else {]]></pattern>
     <message>
-        <id>doubleNegative</id>
+        <id>UnnecessaryElseStatement</id>
         <severity>style</severity>
-        <summary>Double negative is always positive</summary>
+        <summary>Else statement is not necessary.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( \w+ (\||&) \w+ \)]]></pattern>
+    <pattern><![CDATA[\? (true|false) : (true|false)]]></pattern>
     <message>
-        <id>BitwiseOperatorInConditional</id>
+        <id>RedundantConditionalOperator</id>
         <severity>style</severity>
-        <summary>Bitwise operator found in if statement.</summary>
+        <summary>Conditional operator is redundant.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( [^)]+ \) { if \( [^)]+ \) ({[^{}]*(?1)*[^{}]*}) }]]></pattern>
+    <pattern><![CDATA[switch (\([^()]*(?-1)*[^()]*\)) { }]]></pattern>
     <message>
-        <id>CollapsibleIfStatements</id>
+        <id>EmptySwitchStatement</id>
         <severity>style</severity>
-        <summary>These two if statements can be collapsed into one.</summary>
+        <summary>Empty switch statement.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[catch \( [^())]+ \) { }]]></pattern>
+    <pattern><![CDATA[(?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w) ; \1 = [^;]+ ; return \1 ;]]></pattern>
     <message>
-        <id>EmptyCatchStatement</id>
+        <id>RedundantLocalVariable</id>
         <severity>style</severity>
-        <summary>An empty catch statement.</summary>
+        <summary>Variable is returned immediately after its declaration, can be simplified to just return expression.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[do { } while \(]]></pattern>
+    <pattern><![CDATA[for \( ; [^;]+ ; \)]]></pattern>
     <message>
-        <id>EmptyDoWhileStatement</id>
+        <id>ForLoopShouldBeWhileLoop</id>
         <severity>style</severity>
-        <summary>Empty do-while.</summary>
+        <summary>For loop should be written as a while loop.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>normal</tokenlist>
+    <pattern><![CDATA[while (\([^()]*(?-1)*[^()]*\)) { }]]></pattern>
+    <message>
+        <id>EmptyWhileStatement</id>
+        <severity>style</severity>
+        <summary>Empty while statement.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>normal</tokenlist>
+    <pattern><![CDATA[if \( \w+ (\||&) \w+ \)]]></pattern>
+    <message>
+        <id>BitwiseOperatorInConditional</id>
+        <severity>style</severity>
+        <summary>Bitwise operator found in if statement.</summary>
     </message>
 </rule>
 <rule>
@@ -145,7 +172,7 @@
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( [^()]+ \) { }]]></pattern>
+    <pattern><![CDATA[for (\([^()]*(?-1)*[^()]*\)) { }]]></pattern>
     <message>
         <id>EmptyForStatement</id>
         <severity>style</severity>
@@ -154,7 +181,7 @@
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( [^()]+ \) { }]]></pattern>
+    <pattern><![CDATA[if (\([^()]*(?-1)*[^()]*\)) { }]]></pattern>
     <message>
         <id>EmptyIfStatement</id>
         <severity>style</severity>
@@ -163,43 +190,52 @@
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[switch \( [^()]+ \) { }]]></pattern>
+    <pattern><![CDATA[if (\([^()]*(?-1)*[^()]*\)) { return (true|false) ; } else { return (true|false) ; }]]></pattern>
     <message>
-        <id>EmptySwitchStatement</id>
+        <id>RedundantIfStatement</id>
         <severity>style</severity>
-        <summary>Empty switch statement.</summary>
+        <summary>The if statement is redundant.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[while \( [^()]+ \) { }]]></pattern>
+    <pattern><![CDATA[! !]]></pattern>
     <message>
-        <id>EmptyWhileStatement</id>
+        <id>DoubleNegative</id>
         <severity>style</severity>
-        <summary>Empty while statement.</summary>
+        <summary>Double negative is always positive.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[ for \( ; [^;]+ ; \)]]></pattern>
+    <pattern><![CDATA[~ ~]]></pattern>
     <message>
-        <id>ForLoopShouldBeWhileLoop</id>
+        <id>DoubleNegative</id>
         <severity>style</severity>
-        <summary>For loop should be written as a while loop.</summary>
+        <summary>Double negative is always positive.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern>goto</pattern>
+    <pattern><![CDATA[! \( !]]></pattern>
     <message>
-        <id>GotoStatement</id>
+        <id>DoubleNegative</id>
         <severity>style</severity>
-        <summary>Goto considered harmful.</summary>
+        <summary>Double negative is always positive.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>normal</tokenlist>
+    <pattern><![CDATA[~ \( ~]]></pattern>
+    <message>
+        <id>DoubleNegative</id>
+        <severity>style</severity>
+        <summary>Double negative is always positive.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( \w+ != \w+ \) ({[^{}]*(?1)*[^{}]*}) else { (?!if)]]></pattern>
+    <pattern><![CDATA[if \( \w+ != \w+ \) ({[^{}]*(?-1)*[^{}]*}) else { (?!if)]]></pattern>
     <message>
         <id>InvertedLogic</id>
         <severity>style</severity>
@@ -208,7 +244,7 @@
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( ! \w+ \) ({[^{}]*(?1)*[^{}]*}) else { (?!if)]]></pattern>
+    <pattern><![CDATA[if \( ! \w+ \) ({[^{}]*(?-1)*[^{}]*}) else { (?!if)]]></pattern>
     <message>
         <id>InvertedLogic</id>
         <severity>style</severity>
@@ -235,34 +271,43 @@
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[\? (true|false) : (true|false)]]></pattern>
+    <pattern><![CDATA[catch (\([^()]*(?-1)*[^()]*\)) { }]]></pattern>
     <message>
-        <id>RedundantConditionalOperator</id>
+        <id>EmptyCatchStatement</id>
         <severity>style</severity>
-        <summary>Conditional operator is redundant.</summary>
+        <summary>An empty catch statement.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( [^()]+ \) { return (true|false) ; } else { return (true|false) ; }]]></pattern>
+    <pattern><![CDATA[if (\([^()]*(?-1)*[^()]*\)) { assert (\([^()]*(?-1)*[^()]*\)) ; }]]></pattern>
     <message>
-        <id>RedundantIfStatement</id>
+        <id>ConditionalAssert</id>
         <severity>style</severity>
-        <summary>The if statement is redundant.</summary>
+        <summary>The if condition should be included in assert.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[if \( [^()]+ \) { [^{}]* (return|throw|break|continue) [^;]* ; } else {]]></pattern>
+    <pattern><![CDATA[if \( (\w) . empty \( \) \) { for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* \w : \1 \) ({[^{}]*(?-1)*[^{}]*}) }]]></pattern>
     <message>
-        <id>UnnecessaryElseStatement</id>
+        <id>UnnecessaryEmptyCondition</id>
         <severity>style</severity>
-        <summary>Else statement is not necessary.</summary>
+        <summary>Unnecessary check for empty before for range loop.</summary>
     </message>
 </rule>
 <rule>
     <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( \w+ (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = \w+ \[ \1 \] ; }]]></pattern>
+    <pattern><![CDATA[if \( ! (\w) . empty \( \) \) { for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* \w : \1 \) ({[^{}]*(?-1)*[^{}]*}) }]]></pattern>
+    <message>
+        <id>UnnecessaryEmptyCondition</id>
+        <severity>style</severity>
+        <summary>Unnecessary check for empty before for range loop.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = \w+ \[ \1 \] ; }]]></pattern>
     <message>
         <id>useStlAlgorithm</id>
         <severity>style</severity>
@@ -270,8 +315,8 @@
     </message>
 </rule>
 <rule>
-    <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( \w+ (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = \w+ ; }]]></pattern>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = \w+ ; }]]></pattern>
     <message>
         <id>useStlAlgorithm</id>
         <severity>style</severity>
@@ -279,8 +324,8 @@
     </message>
 </rule>
 <rule>
-    <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( \w+ (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (\w+ :: )*\w+ \( \) ; }]]></pattern>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (?:\w+ :: )*\w+ \( \) ; }]]></pattern>
     <message>
         <id>useStlAlgorithm</id>
         <severity>style</severity>
@@ -288,8 +333,8 @@
     </message>
 </rule>
 <rule>
-    <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( \w+ (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (\w+ :: )*\w+ \( \w+ \[ \1 \] \) ; }]]></pattern>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (?:\w+ :: )*\w+ \( \w+ \[ \1 \] \) ; }]]></pattern>
     <message>
         <id>useStlAlgorithm</id>
         <severity>style</severity>
@@ -297,11 +342,65 @@
     </message>
 </rule>
 <rule>
-    <tokenlist>normal</tokenlist>
-    <pattern><![CDATA[for \( \w+ (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (\w+ :: )*\w+ \( \w+ \[ \1 \] , \w+ \[ \1 \] \) ; }]]></pattern>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) = \w+ ; \1 < \w+ ; (\1 \+\+|\+\+ \1|\1 \-\-|\-\- \1) \) { \w+ \[ \1 \] = (?:\w+ :: )*\w+ \( \w+ \[ \1 \] , \w+ \[ \1 \] \) ; }]]></pattern>
     <message>
         <id>useStlAlgorithm</id>
         <severity>style</severity>
         <summary>Considering using std::transform instead.</summary>
     </message>
 </rule>
+<rule>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) : (?:[^()]*(\([^()]*(?-1)*[^()]*\)))*[^)]*\) { (?:(?<idx1>\w+) \+\+|\+\+ (?<idx2>\w+)) ; if (\([^()]*(?-1)*[^()]*\)) { \w+ = \g{idx1}|\g{idx2} ; (?:break ; )?(?:return [^;]*; )?} }]]></pattern>
+    <message>
+        <id>useStlAlgorithm</id>
+        <severity>style</severity>
+        <summary>Considering using std::find or std::find_if instead.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) : (?:[^()]*(\([^()]*(?-1)*[^()]*\)))*[^)]*\) { if (\([^()]*(?-1)*[^()]*\)) { \w+ = (?<idx>\w) ; (?:break ; )?(?:return [^;]*; )?} (?:(\g{idx}) \+\+|\+\+ (\g{idx})) ; }]]></pattern>
+    <message>
+        <id>useStlAlgorithm</id>
+        <severity>style</severity>
+        <summary>Considering using std::find or std::find_if instead.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) : (?:[^()]*(\([^()]*(?-1)*[^()]*\)))*[^)]*\) { (?:(?<idx1>\w+) \+\+|\+\+ (?<idx2>\w+)) ; if (\([^()]*(?-1)*[^()]*\)) { return \g{idx1}|\g{idx2} ; } }]]></pattern>
+    <message>
+        <id>useStlAlgorithm</id>
+        <severity>style</severity>
+        <summary>Considering using std::find or std::find_if instead.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>simple</tokenlist>
+    <pattern><![CDATA[for \( (?:(?:\w+|<|>|::) )*(?:\w+|>)(?: &|\*)* (\w+) : (?:[^()]*(\([^()]*(?-1)*[^()]*\)))*[^)]*\) { if (\([^()]*(?-1)*[^()]*\)) { return (?<idx>\w+) ; } (?:(\g{idx}) \+\+|\+\+ (\g{idx})) ; }]]></pattern>
+    <message>
+        <id>useStlAlgorithm</id>
+        <severity>style</severity>
+        <summary>Considering using std::find or std::find_if instead.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>normal</tokenlist>
+    <pattern><![CDATA[do { } while \(]]></pattern>
+    <message>
+        <id>EmptyDoWhileStatement</id>
+        <severity>style</severity>
+        <summary>Empty do-while.</summary>
+    </message>
+</rule>
+<rule>
+    <tokenlist>normal</tokenlist>
+    <pattern>goto</pattern>
+    <message>
+        <id>GotoStatement</id>
+        <severity>style</severity>
+        <summary>Goto considered harmful.</summary>
+    </message>
+</rule>

dev-requirements.txt

 pfultz2/rocm-recipes
-danmar/cppcheck@681cb7dd909d1bfe41796b7616e43265177b9464 -DHAVE_RULES=1
-ROCm-Developer-Tools/HIP@fc22ef991ce7cb15821c8ccb4f03cdfc3e7e43cf
+danmar/cppcheck@8aa68ee297c2d9ebadf5bcfd00c66ea8d9291e35 -DHAVE_RULES=1
+ROCm-Developer-Tools/HIP@e21df3058728ad8e73708bc99b82e0bdd3509c97
 python/cpython@v3.6.6 -X autotools -H sha256:92aa914572c695c0aeb01b0a214813f414da4b51a371234df514a74761f2bb36
 -f requirements.txt

src/CMakeLists.txt

@@ -5,7 +5,7 @@ include(ROCMPackageConfigHelpers)
 add_library(migraphx 
     auto_contiguous.cpp
     common_subexpression_elimination.cpp
-    constant_propagate.cpp
+    propagate_constant.cpp
     dead_code_elimination.cpp
     eliminate_allocation.cpp
     eliminate_contiguous.cpp
@@ -20,6 +20,7 @@ add_library(migraphx
     program.cpp
     shape.cpp
     schedule.cpp
+    pass_manager.cpp
     simplify_algebra.cpp
     simplify_reshapes.cpp
     opt/memory_coloring.cpp

src/constant_propagate.cpp

-#include <migraphx/constant_propagate.hpp>
-#include <migraphx/program.hpp>
-#include <migraphx/matcher.hpp>
-#include <migraphx/literal.hpp>
-
-namespace migraphx {
-inline namespace MIGRAPHX_INLINE_NS {
-
-struct match_const_add
-{
-    auto matcher() const
-    {
-        return match::name("add")(match::args(match::name("@literal"), match::name("@literal")));
-    }
-
-    void apply(program& p, const match::matcher_result& r) const
-    {
-        auto ins  = r.result;
-        auto arg1 = ins->inputs().at(0)->get_literal();
-        auto arg2 = ins->inputs().at(1)->get_literal();
-
-        auto sum = p.add_literal(transform(arg1, arg2, [](auto x, auto y) { return x + y; }));
-        p.replace_instruction(ins, sum);
-    }
-};
-
-void constant_propagate::apply(program& p) const { match::find_matches(p, match_const_add{}); }
-
-} // namespace MIGRAPHX_INLINE_NS
-} // namespace migraphx

src/eliminate_contiguous.cpp

@@ -9,19 +9,60 @@
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 
-bool try_compute_shape(const operation& op, const std::vector<instruction_ref>& args)
+static bool try_compute_shape(instruction_ref ins, const std::vector<shape>& inputs)
 {
     try
     {
-        compute_shape(op, args);
+        shape new_shape = ins->get_operator().compute_shape(inputs);
+        // If the output shape is a standard shape, no need to try its output
+        if(new_shape.standard())
+        {
+            return true;
+        }
+
+        // if no changes for the shape, the contiguous can also be removed
+        if(new_shape == ins->get_shape())
+        {
+            return true;
+        }
+
+        auto outputs = ins->outputs();
+        // If the current instruction has no output, it means it is the last
+        // instruction and generates a non-standard output shape, and the last
+        // output shape is different from the case with the contiguous operator
+        if(outputs.empty())
+        {
+            return false;
+        }
+
+        for(auto output : outputs)
+        {
+            auto args = output->inputs();
+            std::vector<shape> input_shapes(args.size());
+            std::transform(args.begin(), args.end(), input_shapes.begin(), [&](auto& arg) {
+                return (arg == ins) ? new_shape : arg->get_shape();
+            });
+
+            if(!try_compute_shape(output, input_shapes))
+            {
+                return false;
+            }
+        }
     }
     catch(...)
     {
         return false;
     }
+
     return true;
 }
 
+static bool try_compute_shape(instruction_ref ins, const std::vector<instruction_ref>& args)
+{
+    auto inputs = to_shapes(args);
+    return try_compute_shape(ins, inputs);
+}
+
 void eliminate_contiguous::apply(program& p) const
 {
     for(auto ins : iterator_for(p))
@@ -44,7 +85,7 @@ void eliminate_contiguous::apply(program& p) const
                 auto new_args = args;
                 auto prev     = arg->inputs().front();
                 replace(new_args, arg, prev);
-                if(try_compute_shape(ins->get_operator(), new_args))
+                if(try_compute_shape(ins, new_args))
                 {
                     instruction::replace_argument(ins, arg, prev);
                 }

src/fwd_conv_batchnorm_rewrite.cpp

@@ -63,7 +63,7 @@ void fwd_conv_batchnorm_rewrite::apply(program& p) const
         auto l_weights = p.add_literal({weights.get_shape(), new_weights.data()});
         auto l_bias    = p.add_literal({new_bias.get_shape(), new_bias.data()});
         auto c = p.replace_instruction(conv_ins, conv_op, {conv_ins->inputs()[0], l_weights});
-        auto b = p.insert_instruction(ins, op::broadcast{1, c->get_shape()}, l_bias);
+        auto b = p.insert_instruction(ins, op::broadcast{1, c->get_shape().lens()}, l_bias);
         p.replace_instruction(ins, op::add{}, {c, b});
     }
 }

src/include/migraphx/generate.hpp

@@ -17,7 +17,7 @@ constexpr T normalize(unsigned long z)
         return T(0);
     const auto max     = 32;
     const double range = max / 2; // NOLINT
-    double result      = (z % max) / range;
+    double result      = double(z % max) / range;
     result -= 1;
     return T(result);
 }

src/include/migraphx/instruction.hpp

@@ -24,7 +24,7 @@ struct instruction
 
     instruction(literal l);
 
-    void replace(const shape& r);
+    void replace(operation o);
 
     void recompute_shape();
 
@@ -72,7 +72,9 @@ struct instruction
     static void
     replace(instruction_ref ins, operation o, const shape& r, std::vector<instruction_ref> args);
 
-    argument eval() const;
+    bool can_eval() const;
+
+    argument eval(bool check_eval = true) const;
 
     void finalize(context& ctx);
 
@@ -88,7 +90,8 @@ struct instruction
     // internal
     void replace_argument(instruction_ref old, instruction_ref new_ins);
 
-    private:
+    void replace(const shape& r);
+
     operation op;
     shape result;
     std::vector<instruction_ref> output;

src/include/migraphx/int_divide.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_INT_DIVIDE_HPP
+#define MIGRAPHX_GUARD_RTGLIB_INT_DIVIDE_HPP
+
+#include <migraphx/config.hpp>
+#include <cmath>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+
+template <class R, class T, class U>
+R floor_divide(T x, U y)
+{
+    return R(std::floor(double(x) / double(y)));
+}
+
+template <class R, class T, class U>
+R ceil_divide(T x, U y)
+{
+    return R(std::ceil(double(x) / double(y)));
+}
+
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/include/migraphx/make_signed.hpp

+#ifndef MIGRAPHX_GUARD_RTGLIB_MAKE_SIGNED_HPP
+#define MIGRAPHX_GUARD_RTGLIB_MAKE_SIGNED_HPP
+
+#include <migraphx/config.hpp>
+#include <type_traits>
+#include <utility>
+
+namespace migraphx {
+inline namespace MIGRAPHX_INLINE_NS {
+
+template <class T>
+typename std::conditional_t<std::is_integral<T>{}, std::make_signed<T>, std::enable_if<true, T>>::
+    type
+    make_signed(T x)
+{
+    return x;
+}
+
+} // namespace MIGRAPHX_INLINE_NS
+} // namespace migraphx
+
+#endif

src/include/migraphx/op/abnormal_ops.hpp

@@ -39,6 +39,11 @@ struct undefined
 struct unknown
 {
     std::string op;
+    template <class Self, class F>
+    static auto reflect(Self& self, F f)
+    {
+        return pack(f(self.op, "op"));
+    }
     std::string name() const { return "unknown:" + op; }
     shape compute_shape(std::vector<shape> input) const
     {

src/include/migraphx/op/abs.hpp

@@ -10,6 +10,7 @@
 #include <migraphx/literal.hpp>
 #include <migraphx/shape_for_each.hpp>
 #include <migraphx/config.hpp>
+#include <migraphx/make_signed.hpp>
 #include <cmath>
 #include <utility>
 
@@ -17,9 +18,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct abs : unary
+struct abs : unary<abs>
 {
-    std::string name() const { return "abs"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::abs(make_signed(x)); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/acos.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct acos : unary
+struct acos : unary<acos>
 {
-    std::string name() const { return "acos"; }
+    auto apply() const
+    {
+        return [](auto x) { return std::acos(x); };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/add.hpp

@@ -17,9 +17,12 @@ namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace op {
 
-struct add : binary
+struct add : binary<add>
 {
-    std::string name() const { return "add"; }
+    auto apply() const
+    {
+        return [](auto x, auto y) { return x + y; };
+    }
 };
 
 } // namespace op

src/include/migraphx/op/as_shape.hpp

@@ -36,7 +36,7 @@ struct as_shape
     {
         return {std::move(output_shape), std::move(args.front().data)};
     }
-    int output_alias(const std::vector<shape>&) const { return 0; }
+    std::ptrdiff_t output_alias(const std::vector<shape>&) const { return 0; }
 };
 
 } // namespace op