Roll-up of minimal fixes needed to get external-parameter llama model…

…s working (iree-org#15523) Workarounds for iree-org#15522, iree-org#15520.
nod-ai · Nov 10, 2023 · f66f28f · f66f28f
1 parent 3ef0ea1
commit f66f28f
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 5 deletions.
diff --git a/compiler/src/iree/compiler/Dialect/Stream/IR/StreamTypes.cpp b/compiler/src/iree/compiler/Dialect/Stream/IR/StreamTypes.cpp
@@ -68,6 +68,8 @@ static llvm::cl::opt<bool> clResourceAliasMutableBindings(
     llvm::cl::desc(
         "Fuses bindings that are mutable instead of leaving them split."),
     llvm::cl::init(false));
+// TODO(#15522): Change this to discrete once task system scalability limits
+// are corrected.
 static llvm::cl::opt<IREE::Stream::MemoryModel> clResourceMemoryModel(
     "iree-stream-resource-memory-model",
     llvm::cl::desc("Memory model used for host-device resource memory access."),
@@ -78,7 +80,7 @@ static llvm::cl::opt<IREE::Stream::MemoryModel> clResourceMemoryModel(
         clEnumValN(IREE::Stream::MemoryModel::Discrete, "discrete",
                    "Host and device memory are discrete and cross-access is "
                    "expensive.")),
-    llvm::cl::init(IREE::Stream::MemoryModel::Discrete));
+    llvm::cl::init(IREE::Stream::MemoryModel::Unified));
 
 //===----------------------------------------------------------------------===//
 // custom<ParameterReference>($scope, $key)

diff --git a/runtime/src/iree/io/parameter_index_provider.c b/runtime/src/iree/io/parameter_index_provider.c
@@ -236,11 +236,11 @@ static iree_status_t iree_io_parameter_index_provider_load(
   IREE_RETURN_AND_END_ZONE_IF_ERROR(
       z0, iree_io_parameter_index_provider_resolve(
               provider, device, queue_affinity, source_scope, source_key,
-              target_params.access, &source_entry, &source_file));
+              IREE_HAL_MEMORY_ACCESS_READ, &source_entry, &source_file));
 
   // Validate the parameter range is in-bounds.
   iree_status_t status = iree_io_validate_parameter_range(
-      target_params.access, source_entry, source_offset, length);
+      IREE_HAL_MEMORY_ACCESS_READ, source_entry, source_offset, length);
 
   // Try first to reuse the file backing store directly as a buffer. This only
   // works with specific file types and with specific target usage. The most
@@ -265,7 +265,8 @@ static iree_status_t iree_io_parameter_index_provider_load(
             {
                 .host_allocation =
                     {
-                        .ptr = host_allocation.data,
+                        .ptr = host_allocation.data +
+                               source_entry->storage.file.offset,
                     },
             },
     };

diff --git a/runtime/src/iree/modules/hal/module.c b/runtime/src/iree/modules/hal/module.c
@@ -1212,7 +1212,8 @@ IREE_VM_ABI_EXPORT(iree_hal_module_fence_join,  //
                    CrD, r) {
   iree_host_size_t fence_count = 0;
   iree_hal_fence_t** fences = NULL;
-  IREE_VM_ABI_VLA_STACK_DEREF_OR_NULL(args, a0_count, a0, iree_hal_fence, 32,
+  // TODO(#15520): Rework this to avoid the need for a fixed allocation.
+  IREE_VM_ABI_VLA_STACK_DEREF_OR_NULL(args, a0_count, a0, iree_hal_fence, 349,
                                       &fence_count, &fences);
 
   iree_hal_fence_t* fence = NULL;