GPU CMake: enable the Metal backend by default where the toolchain supports it

ENABLE_METAL defaults to AUTO like the other backends: it is used when the
Metal toolchain compiles MSL 4.1 (macOS 27) and silently skipped otherwise.

Notice that in any case by default we use the CPU, even when Metal is available.

Author: ktf

dependencies/FindO2GPU.cmake

@@ -10,7 +10,7 @@
 # or submit itself to any jurisdiction.
 
 # NOTE!!!! - Whenever this file is changed, move it over to alidist/resources
-# FindO2GPU.cmake Version 23
+# FindO2GPU.cmake Version 25
 
 set(CUDA_COMPUTETARGET_DEFAULT_FULL 80-real;86-real;89-real;120-real;75-virtual)
 set(HIP_AMDGPUTARGET_DEFAULT_FULL gfx906;gfx908)
@@ -21,8 +21,7 @@ if(NOT DEFINED ENABLE_CUDA)
   set(ENABLE_CUDA "AUTO")
 endif()
 if(NOT DEFINED ENABLE_METAL)
-  # OFF rather than AUTO while the backend is unproven
-  set(ENABLE_METAL "OFF")
+  set(ENABLE_METAL "AUTO")
 endif()
 if(NOT DEFINED ENABLE_OPENCL)
   set(ENABLE_OPENCL "AUTO")
@@ -467,15 +466,14 @@ if(ENABLE_METAL)
   # The frameworks are there on every macOS, but the backend needs MSL 4.1, which
   # is macOS 27 and its toolchain, so ask the compiler instead of assuming. A
   # library built as MSL 4.1 also only loads on macOS 27 and later.
-  if(NOT DEFINED GPUCA_METAL_MSL41)
+  # Only a success is cached, so that installing the toolchain is picked up.
+  if(NOT GPUCA_METAL_MSL41)
     set(GPUCA_METAL_PROBE "${CMAKE_CURRENT_BINARY_DIR}/metal_msl41_probe.metal")
     file(WRITE "${GPUCA_METAL_PROBE}" "#include <metal_stdlib>\nkernel void probe(device float* o [[buffer(0)]], uint i [[thread_position_in_grid]]) { o[i] = o[i] * 2.0f; }\n")
     execute_process(COMMAND xcrun -sdk macosx metal -std=metal4.1 -c "${GPUCA_METAL_PROBE}" -o "${GPUCA_METAL_PROBE}.air"
-                    RESULT_VARIABLE GPUCA_METAL_PROBE_RESULT OUTPUT_QUIET ERROR_QUIET)
+                    RESULT_VARIABLE GPUCA_METAL_PROBE_RESULT OUTPUT_QUIET ERROR_VARIABLE GPUCA_METAL_PROBE_ERROR)
     if(GPUCA_METAL_PROBE_RESULT EQUAL 0)
       set(GPUCA_METAL_MSL41 ON CACHE INTERNAL "Metal toolchain compiles MSL 4.1")
-    else()
-      set(GPUCA_METAL_MSL41 OFF CACHE INTERNAL "Metal toolchain compiles MSL 4.1")
     endif()
   endif()
   if(METAL_FRAMEWORK AND COREFOUNDATION_FRAMEWORK AND FOUNDATION_FRAMEWORK AND QUARTZCORE_FRAMEWORK AND GPUCA_METAL_MSL41)
@@ -487,8 +485,9 @@ if(ENABLE_METAL)
     if(NOT METAL_FRAMEWORK OR NOT COREFOUNDATION_FRAMEWORK OR NOT FOUNDATION_FRAMEWORK OR NOT QUARTZCORE_FRAMEWORK)
       message(FATAL_ERROR "Metal frameworks not available")
     endif()
-    message(FATAL_ERROR "The Metal backend needs MSL 4.1: this toolchain rejects -std=metal4.1, and the result would need macOS 27 to run")
+    message(FATAL_ERROR "The Metal backend needs MSL 4.1: this toolchain rejects -std=metal4.1, and the result would need macOS 27 to run\n${GPUCA_METAL_PROBE_ERROR}")
   else()
+    message(STATUS "Metal backend disabled, the toolchain does not compile MSL 4.1: ${GPUCA_METAL_PROBE_ERROR}")
     set(METAL_ENABLED OFF)
   endif()
 endif()