Sync ipex(modify prefetch) (#320)

* fix block_load elems(drop 10%)

* u1 __SYCL_DEPRECATED("u1 is reserved/unsupported") = 1,    // unsigned 1 bit
  s1 __SYCL_DEPRECATED("s1 is reserved/unsupported") = 2,    // signed 1 bit

* save

* update  prefetch gather and scatter 2024.2

* update atomic

* fix comilpe on 2024.1

* conflict

* sync_ipex(update prefetch)

* clang format

* sync ipex 69258612b

* unused var

Author: sunjiweiswift

include/common/core/common.hpp

@@ -227,16 +227,22 @@ enum class atomic_op : uint8_t {
 
 /// xetla dpas argument typ
 enum class argument_type : uint8_t {
-  U1 = 0, // unsigned 1 bit
-  S1 = 1, // signed 1 bit
-  U2 = 2, // unsigned 2 bits
-  S2 = 3, // signed 2 bits
-  U4 = 4, // unsigned 4 bits
-  S4 = 5, // signed 4 bits
-  U8 = 6, // unsigned 8 bits
-  S8 = 7, // signed 8 bits
-  BF16 = 8, // bfloat 16
-  FP16 = 9, // half float
+  Invalid = 0,
+#if __INTEL_LLVM_COMPILER >= 20240200
+  U1 __SYCL_DEPRECATED("u1 is reserved/unsupported") = 1, // unsigned 1 bit
+  S1 __SYCL_DEPRECATED("s1 is reserved/unsupported") = 2, // signed 1 bit
+#else
+  U1 = 1, // unsigned 1 bit
+  S1 = 2, // signed 1 bit
+#endif
+  U2 = 3, // unsigned 2 bits
+  S2 = 4, // signed 2 bits
+  U4 = 5, // unsigned 4 bits
+  S4 = 6, // signed 4 bits
+  U8 = 7, // unsigned 8 bits
+  S8 = 8, // signed 8 bits
+  BF16 = 9, // bfloat 16
+  FP16 = 10, // half float
   TF32 = 12, // tensorfloat 32
   DF = 13, // double (64bits)
   NUM_ARG_TYPES = 14

include/common/core/math_mma.hpp

@@ -33,7 +33,11 @@ namespace detail {
 ///
 template <typename dtype>
 constexpr gpu::xetla::argument_type mma_argument_type() {
+#if __INTEL_LLVM_COMPILER < 20240200
   return gpu::xetla::argument_type::U1;
+#else
+  return gpu::xetla::argument_type::U2;
+#endif
 }
 
 template <>
@@ -72,7 +76,11 @@ constexpr gpu::xetla::argument_type mma_argument_type<fp16>() {
 template <gpu::xetla::argument_type arg_type>
 constexpr __ESIMD_NS::xmx::dpas_argument_type get_argument_type() {
   static_assert(
-      arg_type == gpu::xetla::argument_type::U2 ||
+#if __INTEL_LLVM_COMPILER < 20240200
+      arg_type == gpu::xetla::argument_type::U1 ||
+          arg_type == gpu::xetla::argument_type::S1 ||
+#endif
+          arg_type == gpu::xetla::argument_type::U2 ||
           arg_type == gpu::xetla::argument_type::S2 ||
           arg_type == gpu::xetla::argument_type::U4 ||
           arg_type == gpu::xetla::argument_type::S4 ||
@@ -82,7 +90,14 @@ constexpr __ESIMD_NS::xmx::dpas_argument_type get_argument_type() {
           arg_type == gpu::xetla::argument_type::BF16 ||
           arg_type == gpu::xetla::argument_type::TF32,
       "Unsupported argument type");
+
   switch (arg_type) {
+#if __INTEL_LLVM_COMPILER < 20240200
+    case gpu::xetla::argument_type::U1:
+      return __ESIMD_NS::xmx::dpas_argument_type::u1;
+    case gpu::xetla::argument_type::S1:
+      return __ESIMD_NS::xmx::dpas_argument_type::s1;
+#endif
     case gpu::xetla::argument_type::U2:
       return __ESIMD_NS::xmx::dpas_argument_type::u2;
     case gpu::xetla::argument_type::S2: