(tvm) branch main updated: [Metal] Enable Metal 4 shader compilation (#19595)

[tlopex]

This is an automated email from the ASF dual-hosted git repository.

tlopex pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/tvm.git


The following commit(s) were added to refs/heads/main by this push:
     new b64cb5b2c3 [Metal] Enable Metal 4 shader compilation (#19595)
b64cb5b2c3 is described below

commit b64cb5b2c356109b0543c8c3928f8b935450ba90
Author: Yichen Yan <oralu...@outlook.com>
AuthorDate: Mon Jun 22 10:29:57 2026 +0800

    [Metal] Enable Metal 4 shader compilation (#19595)
    
    Use the latest SDK language version when available so generated MPP
    tensor_ops shaders can compile on M5+ while preserving the existing
    fallback for older SDKs.
---
 src/backend/metal/runtime/metal_module.mm | 27 +++++++++++++++++++++++----
 1 file changed, 23 insertions(+), 4 deletions(-)

diff --git a/src/backend/metal/runtime/metal_module.mm 
b/src/backend/metal/runtime/metal_module.mm
index aff2152a13..a283c27fd8 100644
--- a/src/backend/metal/runtime/metal_module.mm
+++ b/src/backend/metal/runtime/metal_module.mm
@@ -44,12 +44,26 @@
 #include "../../../support/bytes_io.h"
 #include "metal_common.h"
 
+#if (defined(__MAC_OS_X_VERSION_MAX_ALLOWED) && __MAC_OS_X_VERSION_MAX_ALLOWED 
>= 260000) || \
+    (defined(__IPHONE_OS_VERSION_MAX_ALLOWED) && 
__IPHONE_OS_VERSION_MAX_ALLOWED >= 260000)
+#define TVM_METAL_HAS_MSL_4_0 1
+#endif
+
 namespace tvm {
 namespace runtime {
 
 /*! \brief Maximum number of GPU supported in MetalModule. */
 static constexpr const int kMetalMaxNumDevice = 32;
 
+static bool MetalDeviceSupportsMetal4(id<MTLDevice> device) {
+#if defined(TVM_METAL_HAS_MSL_4_0)
+  if (@available(macOS 26.0, iOS 26.0, *)) {
+    return [device supportsFamily:MTLGPUFamilyMetal4];
+  }
+#endif
+  return false;
+}
+
 // Module to support thread-safe multi-GPU execution.
 // The runtime will contain a per-device module table
 // The modules will be lazily loaded
@@ -123,17 +137,22 @@ class MetalModuleNode final : public ffi::ModuleObj {
     const ffi::Bytes& source = (*kernel).second;
 
     if (fmt_ == "metal") {
-      MTLCompileOptions* opts = [MTLCompileOptions alloc];
-      opts.languageVersion = MTLLanguageVersion2_3;
+      MTLCompileOptions* opts = [[MTLCompileOptions alloc] init];
+      MTLLanguageVersion language_version = MTLLanguageVersion2_3;
+#if defined(TVM_METAL_HAS_MSL_4_0)
+      if (MetalDeviceSupportsMetal4(w->devices[device_id])) {
+        language_version = MTLLanguageVersion4_0;
+      }
+#endif
+      opts.languageVersion = language_version;
       opts.fastMathEnabled = YES;
-      // opts = nil;
       // Per-kernel payload is bytes; treat as UTF-8 MSL source.
       std::string source_str(source.data(), source.size());
       lib = [w->devices[device_id]
           newLibraryWithSource:[NSString 
stringWithUTF8String:source_str.c_str()]
                        options:opts
                          error:&err_msg];
-      [opts dealloc];
+      [opts release];
       if (lib == nil) {
         LOG(FATAL) << "Fail to compile metal source:"
                    << [[err_msg localizedDescription] UTF8String];