[CPU] Updated aux GPR count for Load/Store Emitters

openvinotoolkit · Feb 23, 2024 · 82979eb · 82979eb
1 parent c768c0e
commit 82979eb
Show file tree

Hide file tree

Showing 2 changed files with 37 additions and 2 deletions.
diff --git a/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_load_store_emitters.cpp b/src/plugins/intel_cpu/src/emitters/plugin/x64/jit_load_store_emitters.cpp
@@ -62,7 +62,9 @@ size_t store_emitter_params::hash() const {
 }
 
 static int get_aux_regs_as_temp(const size_t byte_size, const bool is_fill = false) {
-    if (!one_of(byte_size, 64u, 32u, 16u))
+    if (one_of(byte_size, 64u, 32u, 16u))
+        return 0;
+    if (one_of(byte_size % 16, 1u, 2u, 3u))
         return 1;
     if (mayiuse(cpu::x64::avx512_core) && is_fill)
         return 1;

diff --git a/src/plugins/intel_cpu/tests/unit/jit_kernel_test.cpp b/src/plugins/intel_cpu/tests/unit/jit_kernel_test.cpp
@@ -324,13 +324,23 @@ TEST(JitKernel, variable_load_and_store) {
     {
         jit_variable_load_store_test_kernel<uint8_t, float> kernel;
         if (mayiuse(cpu_isa_t::avx512_core)) {
-            kernel.test<16, 11, false>();
+            kernel.test<16, 16, false>();
+            kernel.test<16, 15, false>();
+            kernel.test<16, 10, false>();
+            kernel.test<16, 1, false>();
         }
         if (mayiuse(cpu_isa_t::avx2)) {
+            kernel.test<8, 8, false>();
+            kernel.test<8, 7, false>();
+            kernel.test<8, 6, false>();
             kernel.test<8, 5, false>();
+            kernel.test<8, 4, false>();
         }
         if (mayiuse(cpu_isa_t::sse41)) {
+            kernel.test<4, 4, false>();
             kernel.test<4, 3, false>();
+            kernel.test<4, 2, false>();
+            kernel.test<4, 1, false>();
         }
     }
 
@@ -360,6 +370,29 @@ TEST(JitKernel, variable_load_and_store) {
         }
     }
 
+    {
+        jit_variable_load_store_test_kernel<float, uint8_t> kernel;
+        if (mayiuse(cpu_isa_t::avx512_core)) {
+            kernel.test<16, 16, true>();
+            kernel.test<16, 10, true>();
+            kernel.test<16, 2, true>();
+            kernel.test<16, 1, true>();
+        }
+        if (mayiuse(cpu_isa_t::avx2)) {
+            kernel.test<8, 8, true>();
+            kernel.test<8, 7, true>();
+            kernel.test<8, 6, true>();
+            kernel.test<8, 5, true>();
+            kernel.test<8, 4, true>();
+        }
+        if (mayiuse(cpu_isa_t::sse41)) {
+            kernel.test<4, 4, true>();
+            kernel.test<4, 3, true>();
+            kernel.test<4, 2, true>();
+            kernel.test<4, 1, true>();
+        }
+    }
+
     {
         jit_variable_load_store_test_kernel<int32_t, bfloat16_t> kernel;
         if (mayiuse(cpu_isa_t::avx512_core)) {