[arm] add v8-4x8 gemm implement (#4201)

* add v8 4x8 implment. test=devvelop

* fix run error

* change sgemm compute. test=develop

* fix format. test=develop

lite/backends/arm/math/sgemm.cc

@@ -38,9 +38,10 @@ void sgemm(bool is_transA,
            ARMContext* ctx) {
   int hblock = get_hblock(ctx);
   int m_roundup = hblock * ((M + hblock - 1) / hblock);
+  ctx->ExtendWorkspace(m_roundup * K * sizeof(float));
 
-  auto packed_A = static_cast<float*>(
-      TargetMalloc(TargetType::kARM, m_roundup * K * sizeof(float)));
+  auto packed_A = static_cast<float*>(ctx->workspace_data<float>()) +
+                  ctx->llc_size() / sizeof(float);
 
   prepackA(packed_A, A, alpha, lda, 0, M, 0, K, is_transA, ctx);
 
@@ -58,7 +59,6 @@ void sgemm(bool is_transA,
                 is_bias,
                 act_param,
                 ctx);
-  TargetFree(TargetType::kARM, packed_A);
 }
 
 }  // namespace math

lite/tests/math/sgemm_compute_test.cc

@@ -39,7 +39,7 @@ DEFINE_int32(power_mode,
 DEFINE_int32(threads, 1, "threads num");
 DEFINE_int32(warmup, 0, "warmup times");
 DEFINE_int32(repeats, 1, "repeats times");
-DEFINE_bool(basic_test, false, "do all tests");
+DEFINE_bool(basic_test, true, "do all tests");
 DEFINE_bool(check_result, true, "check the result");
 
 DEFINE_int32(M, 512, "gemm: M");