From 8377e52e57c1fb5bcf159aa981f3b25b7bad1cb3 Mon Sep 17 00:00:00 2001
From: Pete Harris <peter.harris@arm.com>
Date: Wed, 17 May 2023 20:14:07 +0100
Subject: [PATCH] Use precise fp with explicit contraction control

VS2019 and older still require strict floating point due to the
older /fp:precise behavior implying contraction, which was
improved in VS2022.
---
 Docs/ChangeLog-4x.md    | 24 +++++++++++++++++-------
 Source/cmake_core.cmake | 13 ++++++++++---
 2 files changed, 27 insertions(+), 10 deletions(-)

diff --git a/Docs/ChangeLog-4x.md b/Docs/ChangeLog-4x.md
index 2a99fac74..7dc1b5beb 100644
--- a/Docs/ChangeLog-4x.md
+++ b/Docs/ChangeLog-4x.md
@@ -14,13 +14,23 @@ clocked at 4.2 GHz, running `astcenc` using AVX2 and 6 threads.
 The 4.5.0 release is a maintenance release with minor fixes and improvements.
 
 * **General:**
-  * **Bug-fix:** Invariant Clang and GCC builds now force `-ffp-model=strict`,
-    which is needed due to recent changes in compiler defaults. These builds
-    will be slower than the previous release, which was not guaranteed to be
-    invariant. Use a non-invariant build to recover lost performance if needed.
-  * **Bug-fix:** Non-invariant Clang and GCC builds now force
-    `-ffp-model=precise`, which is needed due to recent changes in compiler
-    defaults.
+  * **Bug-fix:** Invariant Clang builds now use `-ffp-model=precise` with
+    `-ffp-contract=off` which is needed to restore invariance due to recent
+    changes in compiler defaults.
+  * **Change:** Invariant MSVC builds for VS2022 now use `/fp:precise` instead
+    of `/fp:strict`, which is is now possible because precise no longer implies
+    contraction. This should improve performance for MSVC builds.
+  * **Change:** Non-invariant Clang builds now use `-ffp-model=precise` with
+    `-ffp-contract=on`. This should improve performance on older Clang
+    versions which defaulted to no contraction.
+  * **Change:** Non-invariant MSVC builds for VS2022 now use `/fp:precise`
+    with `/fp:contract`. This should improve performance for MSVC builds.
+  * **Change:** CMake config variables now use an `ASTCENC_` prefix to add a
+    namespace and group options when the library is used in a larger project.
+  * **Change:** CMake config `ASTCENC_NO_INVARIANCE` has been inverted to
+    remove the negated option, and is now `ASTCENC_INVARIANCE` with a default
+    of `ON`. Disablign this option can substantially improve performance, but
+    images can different across platforms and compilers.
 
 <!-- ---------------------------------------------------------------------- -->
 ## 4.4.0
diff --git a/Source/cmake_core.cmake b/Source/cmake_core.cmake
index 42c6f8fe0..2298d25eb 100644
--- a/Source/cmake_core.cmake
+++ b/Source/cmake_core.cmake
@@ -200,16 +200,23 @@ macro(astcenc_set_properties ASTCENC_TARGET_NAME ASTCENC_IS_VENEER)
             PRIVATE
                 ASTCENC_NO_INVARIANCE=1)
 
+        # For Visual Studio prior to 2022 (compiler < 17.0) /fp:precise
+        # For Visual Studio 2022 (compiler >= 17.0) /fp:precise and /fp:contract
         target_compile_options(${ASTCENC_TARGET_NAME}
             PRIVATE
                 $<$<CXX_COMPILER_ID:MSVC>:/fp:precise>
+                $<$<AND:$<CXX_COMPILER_ID:MSVC>,$<VERSION_GREATER_EQUAL:$<CXX_COMPILER_VERSION>,17>>:/fp:contract>
                 $<$<AND:$<PLATFORM_ID:Linux,Darwin>,$<CXX_COMPILER_ID:${CLANG_LIKE}>>:-ffp-model=precise>
                 $<$<PLATFORM_ID:Linux,Darwin>:-ffp-contract=fast>)
     else()
+        # For Visual Studio prior to 2022 (compiler < 17.0) /fp:strict
+        # For Visual Studio 2022 (compiler >= 17.0) /fp:precise
         target_compile_options(${ASTCENC_TARGET_NAME}
             PRIVATE
-                $<$<CXX_COMPILER_ID:MSVC>:/fp:strict>
-                $<$<AND:$<PLATFORM_ID:Linux,Darwin>,$<CXX_COMPILER_ID:${CLANG_LIKE}>>:-ffp-model=strict>)
+                $<$<AND:$<CXX_COMPILER_ID:MSVC>,$<VERSION_LESS:$<CXX_COMPILER_VERSION>,17>>:/fp:strict>
+                $<$<AND:$<CXX_COMPILER_ID:MSVC>,$<VERSION_GREATER_EQUAL:$<CXX_COMPILER_VERSION>,17>>:/fp:precise>
+                $<$<AND:$<PLATFORM_ID:Linux,Darwin>,$<CXX_COMPILER_ID:${CLANG_LIKE}>>:-ffp-model=precise>
+                $<$<PLATFORM_ID:Linux,Darwin>:-ffp-contract=off>)
     endif()
 
     if(${ASTCENC_CLI})
@@ -249,7 +256,7 @@ macro(astcenc_set_properties ASTCENC_TARGET_NAME ASTCENC_IS_VENEER)
 
         # Workaround MSVC codegen bug for NEON builds on VS 2022 17.2 or older
         # https://developercommunity.visualstudio.com/t/inlining-turns-constant-into-register-operand-for/1394798
-        if(CMAKE_CXX_COMPILER_ID MATCHES "MSVC" AND MSVC_VERSION LESS 1933)
+        if((CMAKE_CXX_COMPILER_ID MATCHES "MSVC") AND (MSVC_VERSION LESS 1933))
             target_compile_options(${ASTCENC_TARGET_NAME}
                 PRIVATE
                     $<$<CXX_COMPILER_ID:MSVC>:/d2ssa-cfg-sink->)