Update jit_brgemm_copy_b_emitter to spill only live regs

src/plugins/intel_cpu/src/emitters/snippets/x64/jit_brgemm_copy_b_emitter.cpp

@@ -89,7 +89,10 @@ void jit_brgemm_copy_b_emitter::emit_impl(const std::vector<size_t>& in, const s
     regs_to_spill.emplace(snippets::RegType::gpr, h->rbp.getIdx());
     // Note: abi_param_1 is a default invalid value to check later that the aux reg was allocated properly
     Xbyak::Reg64 aux_reg = abi_param1;
-    utils::init_memory_access_aux_gpr(mem_ptrs_idxs, m_memory_offsets, aux_gpr_idxs, regs_to_spill, aux_reg);
+    const bool is_dynamic_case =
+        std::any_of(m_memory_offsets.cbegin(), m_memory_offsets.cend(), ov::snippets::utils::is_dynamic_value<size_t>);
+    if (is_dynamic_case)
+        aux_reg = utils::init_memory_access_aux_gpr(mem_ptrs_idxs, aux_gpr_idxs, regs_to_spill);
 
     EmitABIRegSpills spill(h);
     spill.preamble(regs_to_spill);

src/plugins/intel_cpu/src/emitters/snippets/x64/jit_brgemm_emitter.cpp

@@ -114,7 +114,10 @@ void jit_brgemm_emitter::emit_call(const std::vector<size_t>& mem_ptrs_idxs) con
     regs_to_spill.emplace(snippets::RegType::gpr, h->rbp.getIdx());
     // Note: abi_param_1 is a default invalid value to check later that the aux reg was allocated properly
     Xbyak::Reg64 aux_reg = abi_param1;
-    utils::init_memory_access_aux_gpr(mem_ptrs_idxs, m_memory_offsets, aux_gpr_idxs, regs_to_spill, aux_reg);
+    const bool is_dynamic_case =
+        std::any_of(m_memory_offsets.cbegin(), m_memory_offsets.cend(), ov::snippets::utils::is_dynamic_value<size_t>);
+    if (std::is_same<T, BrgemmAMXKernelExecutor>() || is_dynamic_case)
+        aux_reg = utils::init_memory_access_aux_gpr(mem_ptrs_idxs, aux_gpr_idxs, regs_to_spill);
     EmitABIRegSpills spill(h);
     spill.preamble(regs_to_spill);
 

src/plugins/intel_cpu/src/emitters/snippets/x64/jit_reg_spill_emitters.cpp

@@ -26,8 +26,6 @@ jit_reg_spill_begin_emitter::jit_reg_spill_begin_emitter(dnnl::impl::cpu::x64::j
     const auto& reg_spill_node = ov::as_type_ptr<snippets::op::RegSpillBegin>(m_reg_spill_begin_expr->get_node());
     OV_CPU_JIT_EMITTER_ASSERT(reg_spill_node, "expects RegSpillBegin expression");
     m_num_spilled = reg_spill_node->get_regs_to_spill().size();
-    // todo: abstract registers were not mapped on physical onces at this point (Kernel emitter does that),
-    //  so we have to declare reg_spiller mutable
     m_abi_reg_spiller = std::make_shared<EmitABIRegSpills>(h);
     in_out_type_ = emitter_in_out_map::gpr_to_gpr;
 }

src/plugins/intel_cpu/src/emitters/snippets/x64/utils.cpp

@@ -59,23 +59,16 @@ Xbyak::Reg64 get_aux_gpr(const std::vector<size_t>& used_gpr_idxs) {
     OV_CPU_JIT_EMITTER_THROW("Failed to allocate aux GPR");
 }
 
-void init_memory_access_aux_gpr(const std::vector<size_t>& mem_ptr_reg_idxs,
-                                const std::vector<size_t>& memory_offsets,
-                                const std::vector<size_t>& aux_gpr_idxs,
-                                std::set<snippets::Reg>& regs_to_spill,
-                                Xbyak::Reg64& aux_reg) {
-    const bool is_dynamic_case =
-        std::any_of(memory_offsets.cbegin(), memory_offsets.cend(), ov::snippets::utils::is_dynamic_value<size_t>);
-    // Note: abi_param_1 is a default invalid value to check later that the aux reg was allocated properly
-    if (is_dynamic_case) {
-        if (!aux_gpr_idxs.empty()) {
-            aux_reg = Xbyak::Reg64(static_cast<int>(aux_gpr_idxs[0]));
-        } else {
-            aux_reg = ov::intel_cpu::utils::get_aux_gpr(mem_ptr_reg_idxs);
-            regs_to_spill.emplace(snippets::RegType::gpr, aux_reg.getIdx());
-        }
+Xbyak::Reg64 init_memory_access_aux_gpr(const std::vector<size_t>& mem_ptr_reg_idxs,
+                                        const std::vector<size_t>& aux_gpr_idxs,
+                                        std::set<snippets::Reg>& regs_to_spill) {
+    if (!aux_gpr_idxs.empty()) {
+        return Xbyak::Reg64(static_cast<int>(aux_gpr_idxs[0]));
     }
-};
+    const auto aux_reg = ov::intel_cpu::utils::get_aux_gpr(mem_ptr_reg_idxs);
+    regs_to_spill.emplace(snippets::RegType::gpr, aux_reg.getIdx());
+    return aux_reg;
+}
 
 void push_ptr_with_runtime_offset_on_stack(dnnl::impl::cpu::x64::jit_generator* h,
                                            size_t stack_offset,

src/plugins/intel_cpu/src/emitters/snippets/x64/utils.hpp

@@ -35,20 +35,15 @@ size_t get_buffer_cluster_id(const ov::snippets::lowered::ExpressionPort& port);
 Xbyak::Reg64 get_aux_gpr(const std::vector<size_t>& used_gpr_idxs);
 
 /**
- * @brief Initializes aux gpr register for dynamic memory access emitters. If any of the `memory_offsets` is dynamic,
- * then try to assign `aux_reg` a register from `aux_gpr_idxs`. If `aux_gpr_idxs` is empty, then choose a register that
- * is not in `mem_ptr_reg_idxs` and add it to `regs_to_spill`.
+ * @brief Returns aux gpr register for dynamic memory access emitters. Returns a register from `aux_gpr_idxs`.
+ * If it's empty, then choose a register that is not in `mem_ptr_reg_idxs` and add it to `regs_to_spill`.
  * @param mem_ptr_reg_idxs register indexes reserved to store memory pointers in this emitter
- * @param memory_offsets memory offsets, could be dynamic
  * @param aux_gpr_idxs pool of available gp register indexes
  * @param regs_to_spill set of live registers to be spilled before ABI call
- * @param aux_reg auxiliary register that should be initialized
  */
-void init_memory_access_aux_gpr(const std::vector<size_t>& mem_ptr_reg_idxs,
-                                const std::vector<size_t>& memory_offsets,
-                                const std::vector<size_t>& aux_gpr_idxs,
-                                std::set<snippets::Reg>& regs_to_spill,
-                                Xbyak::Reg64& aux_reg);
+Xbyak::Reg64 init_memory_access_aux_gpr(const std::vector<size_t>& mem_ptr_reg_idxs,
+                                        const std::vector<size_t>& aux_gpr_idxs,
+                                        std::set<snippets::Reg>& regs_to_spill);
 
 /**
  * @brief Push data pointer on stack adding offset. The offset is taken from runtime params `abi_param1`