Fix segfault and other fixes

sshlyapn · Oct 17, 2024 · 53cabec · 53cabec
1 parent f5e6784
commit 53cabec
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/src/plugins/intel_gpu/src/graph/primitive_inst.cpp b/src/plugins/intel_gpu/src/graph/primitive_inst.cpp
@@ -370,9 +370,9 @@ void primitive_inst::update_shape() {
                 update_state_layout(*scales_state, new_scales_layout, 1);
 
                 if (compressed_cache_variable->has_zp_state()) {
-                    auto zp_state = compressed_cache_variable->get_compression_zp_state();
+                    auto scales_state = compressed_cache_variable->get_compression_zp_state();
                     auto new_zp_layout = compressed_cache_variable->get_compression_zp_state()->get_layout();
-                    update_state_layout(*zp_state, new_zp_layout, 2);
+                    update_state_layout(*scales_state, new_zp_layout, 2);
 
                     GPU_DEBUG_TRACE_DETAIL << "Set new zp layout " << new_zp_layout.to_short_string() << "\n";
                     GPU_DEBUG_TRACE_DETAIL << "states[2] " << _impl_params->state_layouts[2].to_short_string() << "\n";
@@ -1414,7 +1414,7 @@ void primitive_inst::do_runtime_in_place_kv_cache() {
                 const auto sequence_axis = 2;
                 kv_cache_inst::update_pad(past_scale_layout, max_pad, sequence_axis);
 
-                if (_impl_params->input_layouts.size() >= 4) {
+                if (_impl_params->input_layouts.size() > 4) {
                     auto& past_zp_layout = _impl_params->input_layouts[4];
                     const auto sequence_axis = 2;
                     kv_cache_inst::update_pad(past_zp_layout, max_pad, sequence_axis);

diff --git a/src/plugins/intel_gpu/src/kernel_selector/cl_kernels/sdpa_ref.cl b/src/plugins/intel_gpu/src/kernel_selector/cl_kernels/sdpa_ref.cl
@@ -131,7 +131,7 @@ KERNEL(sdpa_ref)(
     const __global VALUE_COMPRESSION_SCALE_TYPE* val_scale,
 #if HAS_KV_CACHE_ZP_INPUT
     const __global KEY_COMPRESSION_ZP_TYPE* key_zp,
-    const __global VALUE_COMPRESSION_SCALE_TYPE* val_zp,
+    const __global VALUE_COMPRESSION_ZP_TYPE* val_zp,
 #endif
 #endif
 #ifdef BEAM_TABLE_TYPE