Test google/pthreadpool fix

Author: GregoryComer

.ci/scripts/test_lora.sh

@@ -139,7 +139,8 @@ Okay, so I need to calculate 15% of 80."
 EXPECTED_QUANT_LORA_PREFIX="
 <|im_start|>user Calculate 15% of 80?<|im_end|><|im_start|>assistant
 To calculate 15% of 80, we can multiply 80 by 15/100.
-So, 15% of 80 is equal to (80 * 15) / 100 = 1200 / 100 = 12.
+80 * 15/100 = 12.
+So, 15% of 80 is 12.
 #### 12
 The answer is: 12<|im_end|>"
 

.gitmodules

@@ -24,7 +24,7 @@
 	url = https://github.com/pytorch/cpuinfo.git
 [submodule "backends/xnnpack/third-party/pthreadpool"]
 	path = backends/xnnpack/third-party/pthreadpool
-	url = https://github.com/google/pthreadpool.git
+	url = https://github.com/gregorycomer/pthreadpool.git
 [submodule "extension/llm/tokenizers"]
 	path = extension/llm/tokenizers
 	url = https://github.com/meta-pytorch/tokenizers.git

backends/xnnpack/CMakeLists.txt

@@ -154,12 +154,16 @@ install(
           xnnpack-normalization
           xnnpack-operators
           xnnpack-operator-run
+          xnnpack-operator-delete
           xnnpack-operator-utils
           xnnpack-pack-lh
           xnnpack-packing
           xnnpack-sanitizers
           xnnpack-subgraph
           xnnpack-datatype
+          xnnpack-fingerprint-id
+          xnnpack-fingerprint-cache
+          xnnpack-fingerprint-check
           xnnpack-reference-ukernels
           xnnpack-logging
   EXPORT ExecuTorchTargets

backends/xnnpack/third-party/XNNPACK

@@ -41,7 +41,10 @@ def define_xnnpack():
             "XNNPACK/src/memory.c",
             "XNNPACK/src/mutex.c",
             "XNNPACK/src/normalization.c",
+            "XNNPACK/src/operator-delete.c",
             "XNNPACK/src/operator-utils.c",
+            "XNNPACK/src/operators/fingerprint_cache.c",
+            "XNNPACK/src/operators/fingerprint_id.c",
             "XNNPACK/src/reference/packing.cc",
         ],
         headers = get_xnnpack_headers(),
@@ -1039,7 +1042,7 @@ def define_xnnpack():
     native.cxx_library(
         name = "ukernels_avx512vnnigfni",
         srcs = select({
-            "DEFAULT": prod_srcs_for_arch_wrapper("avx512vnnifgni"),
+            "DEFAULT": prod_srcs_for_arch_wrapper("avx512vnnigfni"),
             "ovr_config//cpu:arm32": DEFAULT_DUMMY_SRC,
             "ovr_config//cpu:arm64": DEFAULT_DUMMY_SRC,
         }),
@@ -1068,6 +1071,7 @@ def define_xnnpack():
         "-mavxvnni",
         "-mf16c",
         "-mfma",
+        "-mgfni",
     ]
 
     # @lint-ignore BUCKLINT: native and fb_native are explicitly forbidden in fbcode.
@@ -1172,6 +1176,14 @@ def define_xnnpack():
             # "-DXNN_ENABLE_DWCONV_MULTIPLASS=0",
             "-DXNN_ENABLE_ARM_I8MM=1",
             "-DXNN_ENABLE_ARM_FP16_VECTOR=1",
+            "-DXNN_ENABLE_SSE=1",
+            "-DXNN_ENABLE_SSE2=1",
+            "-DXNN_ENABLE_SSSE3=1",
+            "-DXNN_ENABLE_SSE41=1",
+            "-DXNN_ENABLE_AVX=1",
+            "-DXNN_ENABLE_F16C=1",
+            "-DXNN_ENABLE_FMA3=1",
+            "-DXNN_ENABLE_AVX2=1",
             "-DXNN_ENABLE_AVX512F=1",
             "-DXNN_ENABLE_AVX512SKX=1",
             "-DXNN_ENABLE_AVX512VNNI=1",