[CPU] Add FP8 Bmm support (#9744)

Co-authored-by: Fan Yin <1106310035@qq.com>

sgl-kernel/csrc/cpu/gemm.h

@@ -232,6 +232,7 @@ void tinygemm_kernel(
     int64_t ldc,
     bool brg);
 
+// block quantization
 template <typename scalar_t>
 void tinygemm_kernel(
     const scalar_t* __restrict__ A,
@@ -250,6 +251,23 @@ void tinygemm_kernel(
     int64_t block_size_K,
     bool do_unpack = true);
 
+// per tensor quantization
+template <typename scalar_t>
+void tinygemm_kernel(
+    const scalar_t* __restrict__ A,
+    const at::Float8_e4m3fn* __restrict__ B,
+    scalar_t* __restrict__ C,
+    scalar_t* __restrict__ Btmp,
+    float* __restrict__ Ctmp,
+    float scale,
+    int64_t M,
+    int64_t N,
+    int64_t K,
+    int64_t lda,
+    int64_t ldb,
+    int64_t ldc,
+    bool brg);
+
 template <typename scalar_t>
 void tinygemm_kernel(
     scalar_t* C,