ISPC: Support generation for multiple instruction sets

16 files changed, 305 insertions, 2 deletions

diff --git a/Tests/ISPC/CMakeLists.txt b/Tests/ISPC/CMakeLists.txt
index ca5a79b..c13271a 100644
--- a/Tests/ISPC/CMakeLists.txt
+++ b/Tests/ISPC/CMakeLists.txt
@@ -6,7 +6,10 @@ macro (add_ispc_test_macro name)
     PROPERTY LABELS "ISPC")
 endmacro ()
 
+add_ispc_test_macro(ISPC.ChainedStaticLibraries ISPCChainedStaticLibraries)
 add_ispc_test_macro(ISPC.Defines ISPCDefines)
+add_ispc_test_macro(ISPC.DynamicLibrary ISPCDynamicLibrary)
+add_ispc_test_macro(ISPC.ObjectGenex ISPCObjectGenex)
 add_ispc_test_macro(ISPC.ObjectLibrary ISPCObjectLibrary)
 add_ispc_test_macro(ISPC.ResponseAndDefine ISPCResponseAndDefine)
 add_ispc_test_macro(ISPC.StaticLibrary ISPCStaticLibrary)
diff --git a/Tests/ISPC/ChainedStaticLibraries/CMakeLists.txt b/Tests/ISPC/ChainedStaticLibraries/CMakeLists.txt
new file mode 100644
index 0000000..bf0b57e
--- /dev/null
+++ b/Tests/ISPC/ChainedStaticLibraries/CMakeLists.txt
@@ -0,0 +1,22 @@
+
+cmake_minimum_required(VERSION 3.18)
+project(ISPCChainedStaticLibraries CXX ISPC)
+
+if(CMAKE_SIZEOF_VOID_P EQUAL 4)
+  set(CMAKE_ISPC_FLAGS "--arch=x86")
+endif()
+
+
+add_library(ispc_objects1 STATIC extra.ispc extra.cxx)
+add_library(ispc_objects2 STATIC simple.ispc)
+
+set_target_properties(ispc_objects1 PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(ispc_objects1 PROPERTIES ISPC_INSTRUCTION_SETS "avx1-i32x16;avx2-i32x4")
+
+set_target_properties(ispc_objects2 PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(ispc_objects2 PROPERTIES ISPC_INSTRUCTION_SETS "avx1-i32x16")
+
+target_link_libraries(ispc_objects2 PRIVATE ispc_objects1)
+
+add_executable(ISPCChainedStaticLibraries main.cxx)
+target_link_libraries(ISPCChainedStaticLibraries PUBLIC ispc_objects2)
diff --git a/Tests/ISPC/ChainedStaticLibraries/extra.cxx b/Tests/ISPC/ChainedStaticLibraries/extra.cxx
new file mode 100644
index 0000000..88ef3a7
--- /dev/null
+++ b/Tests/ISPC/ChainedStaticLibraries/extra.cxx
@@ -0,0 +1,17 @@
+#include <stdio.h>
+
+#include "extra.ispc.h"
+
+int extra()
+{
+  float vin[16], vout[16];
+  for (int i = 0; i < 16; ++i)
+    vin[i] = i;
+
+  ispc::extra(vin, vout, 16);
+
+  for (int i = 0; i < 16; ++i)
+    printf("%d: extra(%f) = %f\n", i, vin[i], vout[i]);
+
+  return 0;
+}
diff --git a/Tests/ISPC/ChainedStaticLibraries/extra.ispc b/Tests/ISPC/ChainedStaticLibraries/extra.ispc
new file mode 100644
index 0000000..5a4a442
--- /dev/null
+++ b/Tests/ISPC/ChainedStaticLibraries/extra.ispc
@@ -0,0 +1,12 @@
+
+export void extra(uniform float vin[], uniform float vout[],
+                   uniform int count) {
+    foreach (index = 0 ... count) {
+        float v = vin[index];
+        if (v < 3.)
+            v = v * v;
+        else
+            v = sqrt(v);
+        vout[index] = v;
+    }
+}
diff --git a/Tests/ISPC/ChainedStaticLibraries/main.cxx b/Tests/ISPC/ChainedStaticLibraries/main.cxx
new file mode 100644
index 0000000..4f1c9be
--- /dev/null
+++ b/Tests/ISPC/ChainedStaticLibraries/main.cxx
@@ -0,0 +1,15 @@
+#include <stdio.h>
+
+#include "simple.ispc.h"
+
+int main()
+{
+  float vin[16], vout[16];
+  for (int i = 0; i < 16; ++i)
+    vin[i] = i;
+
+  ispc::simple(vin, vout, 16);
+
+  for (int i = 0; i < 16; ++i)
+    printf("%d: simple(%f) = %f\n", i, vin[i], vout[i]);
+}
diff --git a/Tests/ISPC/ChainedStaticLibraries/simple.ispc b/Tests/ISPC/ChainedStaticLibraries/simple.ispc
new file mode 100644
index 0000000..70cb588
--- /dev/null
+++ b/Tests/ISPC/ChainedStaticLibraries/simple.ispc
@@ -0,0 +1,12 @@
+
+export void simple(uniform float vin[], uniform float vout[],
+                   uniform int count) {
+    foreach (index = 0 ... count) {
+        float v = vin[index];
+        if (v < 3.)
+            v = v * v;
+        else
+            v = sqrt(v);
+        vout[index] = v;
+    }
+}
diff --git a/Tests/ISPC/Defines/CMakeLists.txt b/Tests/ISPC/Defines/CMakeLists.txt
index 5155106..7645804 100644
--- a/Tests/ISPC/Defines/CMakeLists.txt
+++ b/Tests/ISPC/Defines/CMakeLists.txt
@@ -1,6 +1,7 @@
 cmake_minimum_required(VERSION 3.18)
 project(ISPCDefines CXX ISPC)
 
+set(CMAKE_ISPC_INSTRUCTION_SETS "sse2-i32x4;sse4-i16x8;avx1-i32x16;avx2-i32x4;avx512knl-i32x16;avx512skx-i32x8")
 set(CMAKE_ISPC_FLAGS -DM_PI=3.1415926535f)
 add_compile_definitions([==[STRUCT_DEFINE=struct{uniform int a]==])
 
diff --git a/Tests/ISPC/DynamicLibrary/CMakeLists.txt b/Tests/ISPC/DynamicLibrary/CMakeLists.txt
new file mode 100644
index 0000000..59f875e
--- /dev/null
+++ b/Tests/ISPC/DynamicLibrary/CMakeLists.txt
@@ -0,0 +1,20 @@
+
+cmake_minimum_required(VERSION 3.18)
+project(ISPCDynamicLibrary CXX ISPC)
+
+if(CMAKE_SIZEOF_VOID_P EQUAL 4)
+  set(CMAKE_ISPC_FLAGS "--arch=x86")
+endif()
+
+
+add_library(ispc_objects1 STATIC extra.ispc extra.cxx)
+add_library(ispc_objects2 SHARED simple.ispc)
+
+set_target_properties(ispc_objects1 PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(ispc_objects1 PROPERTIES ISPC_INSTRUCTION_SETS "avx1-i32x16;avx2-i32x4")
+set_target_properties(ispc_objects2 PROPERTIES ISPC_INSTRUCTION_SETS "avx1-i32x16")
+
+target_link_libraries(ispc_objects2 PRIVATE ispc_objects1)
+
+add_executable(ISPCDynamicLibrary main.cxx)
+target_link_libraries(ISPCDynamicLibrary PUBLIC ispc_objects2)
diff --git a/Tests/ISPC/DynamicLibrary/extra.cxx b/Tests/ISPC/DynamicLibrary/extra.cxx
new file mode 100644
index 0000000..88ef3a7
--- /dev/null
+++ b/Tests/ISPC/DynamicLibrary/extra.cxx
@@ -0,0 +1,17 @@
+#include <stdio.h>
+
+#include "extra.ispc.h"
+
+int extra()
+{
+  float vin[16], vout[16];
+  for (int i = 0; i < 16; ++i)
+    vin[i] = i;
+
+  ispc::extra(vin, vout, 16);
+
+  for (int i = 0; i < 16; ++i)
+    printf("%d: extra(%f) = %f\n", i, vin[i], vout[i]);
+
+  return 0;
+}
diff --git a/Tests/ISPC/DynamicLibrary/extra.ispc b/Tests/ISPC/DynamicLibrary/extra.ispc
new file mode 100644
index 0000000..5a4a442
--- /dev/null
+++ b/Tests/ISPC/DynamicLibrary/extra.ispc
@@ -0,0 +1,12 @@
+
+export void extra(uniform float vin[], uniform float vout[],
+                   uniform int count) {
+    foreach (index = 0 ... count) {
+        float v = vin[index];
+        if (v < 3.)
+            v = v * v;
+        else
+            v = sqrt(v);
+        vout[index] = v;
+    }
+}
diff --git a/Tests/ISPC/DynamicLibrary/main.cxx b/Tests/ISPC/DynamicLibrary/main.cxx
new file mode 100644
index 0000000..4f1c9be
--- /dev/null
+++ b/Tests/ISPC/DynamicLibrary/main.cxx
@@ -0,0 +1,15 @@
+#include <stdio.h>
+
+#include "simple.ispc.h"
+
+int main()
+{
+  float vin[16], vout[16];
+  for (int i = 0; i < 16; ++i)
+    vin[i] = i;
+
+  ispc::simple(vin, vout, 16);
+
+  for (int i = 0; i < 16; ++i)
+    printf("%d: simple(%f) = %f\n", i, vin[i], vout[i]);
+}
diff --git a/Tests/ISPC/DynamicLibrary/simple.ispc b/Tests/ISPC/DynamicLibrary/simple.ispc
new file mode 100644
index 0000000..70cb588
--- /dev/null
+++ b/Tests/ISPC/DynamicLibrary/simple.ispc
@@ -0,0 +1,12 @@
+
+export void simple(uniform float vin[], uniform float vout[],
+                   uniform int count) {
+    foreach (index = 0 ... count) {
+        float v = vin[index];
+        if (v < 3.)
+            v = v * v;
+        else
+            v = sqrt(v);
+        vout[index] = v;
+    }
+}
diff --git a/Tests/ISPC/ObjectGenex/CMakeLists.txt b/Tests/ISPC/ObjectGenex/CMakeLists.txt
new file mode 100644
index 0000000..bc0cbf6
--- /dev/null
+++ b/Tests/ISPC/ObjectGenex/CMakeLists.txt
@@ -0,0 +1,45 @@
+cmake_minimum_required(VERSION 3.18)
+project(ISPCObjectGenex CXX ISPC)
+
+set(CMAKE_ISPC_INSTRUCTION_SETS "sse2-i32x4;sse4-i16x8;avx1-i32x16;avx2-i32x4;avx512knl-i32x16;avx512skx-i32x8")
+
+add_library(ispc_objects OBJECT
+  simple.ispc
+  )
+target_compile_definitions(ispc_objects PRIVATE
+  $<$<COMPILE_LANG_AND_ID:ISPC,Intel>:M_PI=3.1415926535f>
+)
+set_target_properties(ispc_objects PROPERTIES POSITION_INDEPENDENT_CODE ON)
+if(CMAKE_SIZEOF_VOID_P EQUAL 4)
+  set_source_files_properties(simple.ispc PROPERTIES COMPILE_OPTIONS "--arch=x86")
+endif()
+
+
+#Test ObjectFiles with file(GENERATE)
+file(GENERATE
+     OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/gen_$<LOWER_CASE:$<CONFIG>/>path_to_objs.h
+     CONTENT [[
+#ifndef path_to_objs
+#define path_to_objs
+
+#include <string>
+
+static std::string obj_paths = "$<TARGET_OBJECTS:ispc_objects>";
+
+#endif
+
+]]
+)
+
+
+add_executable(ISPCObjectGenex main.cxx)
+add_dependencies(ISPCObjectGenex ispc_objects)
+
+list(LENGTH CMAKE_ISPC_INSTRUCTION_SETS numberOfTargets)
+math(EXPR numberOfTargets "${numberOfTargets}+1")
+target_compile_definitions(ISPCObjectGenex PRIVATE
+    "ExpectedISPCObjects=${numberOfTargets}"
+    "CONFIG_TYPE=gen_$<LOWER_CASE:$<CONFIG>>"
+    )
+target_include_directories(ISPCObjectGenex PRIVATE ${CMAKE_CURRENT_BINARY_DIR} )
+target_compile_features(ISPCObjectGenex PRIVATE cxx_std_11)
diff --git a/Tests/ISPC/ObjectGenex/main.cxx b/Tests/ISPC/ObjectGenex/main.cxx
new file mode 100644
index 0000000..143e74e
--- /dev/null
+++ b/Tests/ISPC/ObjectGenex/main.cxx
@@ -0,0 +1,87 @@
+#include <stdio.h>
+
+/*
+  Define GENERATED_HEADER macro to allow c++ files to include headers
+  generated based on different configuration types.
+*/
+
+/* clang-format off */
+#define GENERATED_HEADER(x) GENERATED_HEADER0(CONFIG_TYPE/x)
+/* clang-format on */
+#define GENERATED_HEADER0(x) GENERATED_HEADER1(x)
+#define GENERATED_HEADER1(x) <x>
+
+#include GENERATED_HEADER(path_to_objs.h)
+
+#include <vector>
+std::vector<std::string> expandList(std::string const& arg)
+{
+  std::vector<std::string> output;
+  // If argument is empty or no `;` just copy the current string
+  if (arg.empty() || arg.find(';') == std::string::npos) {
+    output.emplace_back(arg);
+    return output;
+  }
+
+  std::string newArg;
+  // Break the string at non-escaped semicolons not nested in [].
+  int squareNesting = 0;
+  auto last = arg.begin();
+  auto const cend = arg.end();
+  for (auto c = last; c != cend; ++c) {
+    switch (*c) {
+      case '\\': {
+        // We only want to allow escaping of semicolons.  Other
+        // escapes should not be processed here.
+        auto cnext = c + 1;
+        if ((cnext != cend) && *cnext == ';') {
+          newArg.append(last, c);
+          // Skip over the escape character
+          last = cnext;
+          c = cnext;
+        }
+      } break;
+      case '[': {
+        ++squareNesting;
+      } break;
+      case ']': {
+        --squareNesting;
+      } break;
+      case ';': {
+        // Break the string here if we are not nested inside square
+        // brackets.
+        if (squareNesting == 0) {
+          newArg.append(last, c);
+          // Skip over the semicolon
+          last = c + 1;
+          if (!newArg.empty()) {
+            // Add the last argument if the string is not empty.
+            output.push_back(newArg);
+            newArg.clear();
+          }
+        }
+      } break;
+      default: {
+        // Just append this character.
+      } break;
+    }
+  }
+  newArg.append(last, cend);
+  if (!newArg.empty()) {
+    // Add the last argument if the string is not empty.
+    output.push_back(std::move(newArg));
+  }
+
+  return output;
+}
+
+int main()
+{
+  // determine that the number of object files specified in obj_paths
+  // is equal to the number of arch's
+
+  std::vector<std::string> paths = expandList(obj_paths);
+  const bool correctSize = (paths.size() == ExpectedISPCObjects);
+
+  return (correctSize) ? 0 : 1;
+}
diff --git a/Tests/ISPC/ObjectGenex/simple.ispc b/Tests/ISPC/ObjectGenex/simple.ispc
new file mode 100644
index 0000000..a76b76c
--- /dev/null
+++ b/Tests/ISPC/ObjectGenex/simple.ispc
@@ -0,0 +1,12 @@
+
+export void simple(uniform float vin[], uniform float vout[],
+                   uniform int count) {
+    foreach (index = 0 ... count) {
+        float v = vin[index];
+        if (v < M_PI)
+            v = v * v;
+        else
+            v = sqrt(v);
+        vout[index] = v;
+    }
+}
diff --git a/Tests/ISPC/ObjectLibrary/CMakeLists.txt b/Tests/ISPC/ObjectLibrary/CMakeLists.txt
index 333ad66..4767d7e 100644
--- a/Tests/ISPC/ObjectLibrary/CMakeLists.txt
+++ b/Tests/ISPC/ObjectLibrary/CMakeLists.txt
@@ -7,11 +7,12 @@ if(CMAKE_SIZEOF_VOID_P EQUAL 4)
   set(CMAKE_ISPC_FLAGS "--arch=x86")
 endif()
 
-add_library(ispc_objects OBJECT simple.ispc extra.ispc)
 
-target_compile_options(ispc_objects PRIVATE "$<$<COMPILE_LANGUAGE:ISPC>:--target=sse2-i32x4>")
+add_library(ispc_objects OBJECT simple.ispc extra.ispc)
 
 set_target_properties(ispc_objects PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(ispc_objects PROPERTIES ISPC_INSTRUCTION_SETS "sse2-i32x4;sse4-i8x16")
+
 
 add_executable(ISPCObjectLibrary main.cxx extra.cxx)
 target_link_libraries(ISPCObjectLibrary PRIVATE ispc_objects)