Add xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w

src/configs/gemm-config.c

@@ -5042,7 +5042,7 @@ static void init_qs8_qc4w_gemm_config(void) {
         qs8_qc4w_gemm_config.init.qs8_qc8w = xnn_init_qs8_qc8w_conv_minmax_fp32_scalar_params;
         // TODO(fbarchard): generate avx512/avx2 packw instead of scalar in future
         qs8_qc4w_gemm_config.pack_gemm_goi = (xnn_packw_gemm_goi_ukernel_fn) xnn_qs8_to_qu8_qc4uw_packw_gemm_goi_ukernel_x16c8__scalar;
-        // TODO(fbarchard): add xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w in future
+        qs8_qc4w_gemm_config.pack_gemm_gio = (xnn_packw_gemm_gio_ukernel_fn) xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w;
         qs8_qc4w_gemm_config.pack_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.packed_stride_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.planes = 2;
@@ -5059,7 +5059,7 @@ static void init_qs8_qc4w_gemm_config(void) {
         qs8_qc4w_gemm_config.init.qs8_qc8w = xnn_init_qs8_qc8w_conv_minmax_fp32_scalar_params;
         // TODO(fbarchard): generate avx2 packw instead of sse2 in future
         qs8_qc4w_gemm_config.pack_gemm_goi = (xnn_packw_gemm_goi_ukernel_fn) xnn_qs8_to_qu8_qc4uw_packw_gemm_goi_ukernel_x8c8__sse2;
-        // TODO(fbarchard): add xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w in future
+        qs8_qc4w_gemm_config.pack_gemm_gio = (xnn_packw_gemm_gio_ukernel_fn) xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w;
         qs8_qc4w_gemm_config.pack_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.packed_stride_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.planes = 2;
@@ -5076,7 +5076,7 @@ static void init_qs8_qc4w_gemm_config(void) {
         qs8_qc4w_gemm_config.init.qs8_qc8w = xnn_init_qs8_qc8w_conv_minmax_fp32_scalar_params;
         // TODO(fbarchard): generate avx2 packw instead of sse2 in future
         qs8_qc4w_gemm_config.pack_gemm_goi = (xnn_packw_gemm_goi_ukernel_fn) xnn_qs8_to_qu8_qc4uw_packw_gemm_goi_ukernel_x8c8__sse2;
-        // TODO(fbarchard): add xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w in future
+        qs8_qc4w_gemm_config.pack_gemm_gio = (xnn_packw_gemm_gio_ukernel_fn) xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w;
         qs8_qc4w_gemm_config.pack_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.packed_stride_weights_and_biases = NULL;
         qs8_qc4w_gemm_config.planes = 2;

src/reference/packing.cc

@@ -1622,11 +1622,12 @@ void xnn_pack_qs8_qc4w_gemm_gio_w(
 // Same as qc4w but unsigned 4 bit output
 // Applies kv ^ 0x88 to convert int4 to uint4
 // Does not multiply bias by 16
-void xnn_pack_qs8_qc4uw_gemm_gio_w(
-    size_t g, size_t nc, size_t kc, size_t nr, size_t kr, size_t sr,
-    size_t k_stride, const uint8_t* k, const int32_t* b, const float* scale,
-    void* packed_weights, size_t extra_bytes,
-    const struct xnn_qs8_qc4w_packing_params* params) {
+static void pack_qs8_qc4uw_gemm_gio_w(size_t g, size_t nc, size_t kc, size_t nr,
+                                      size_t kr, size_t sr, size_t k_stride,
+                                      const uint8_t* k, const int32_t* b,
+                                      const float* scale, void* packed_weights,
+                                      size_t extra_bytes, uint32_t izp,
+                                      uint32_t kernel_zero_point) {
   assert(g != 0);
   assert(nc != 0);
   assert(kc != 0);
@@ -1635,12 +1636,9 @@ void xnn_pack_qs8_qc4uw_gemm_gio_w(
   assert(sr >= 1 && sr <= 16);
   assert(k != nullptr);
   assert(packed_weights != nullptr);
-  assert(params != nullptr);
-  assert(params->kernel_zero_point == 8 || params->kernel_zero_point == 0);
+  assert(kernel_zero_point == 8 || kernel_zero_point == 0);
 
   const size_t skr = sr * kr;
-  const uint32_t izp = (uint32_t)params->input_zero_point;
-  const uint32_t kernel_zero_point = (uint32_t)params->kernel_zero_point;
   do {
     size_t nr_block_start = 0;
     do {
@@ -1717,6 +1715,27 @@ void xnn_pack_qs8_qc4uw_gemm_gio_w(
   } while (--g != 0);
 }
 
+void xnn_pack_qs8_qc4uw_gemm_gio_w(
+    size_t g, size_t nc, size_t kc, size_t nr, size_t kr, size_t sr,
+    size_t k_stride, const uint8_t* k, const int32_t* b, const float* scale,
+    void* packed_weights, size_t extra_bytes,
+    const struct xnn_qs8_qc4w_packing_params* params) {
+  assert(params != nullptr);
+  pack_qs8_qc4uw_gemm_gio_w(
+      g, nc, kc, nr, kr, sr, k_stride, k, b, scale, packed_weights, extra_bytes,
+      params->input_zero_point, params->kernel_zero_point);
+}
+// For qs8_qc4w madd
+void xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w(
+    size_t g, size_t nc, size_t kc, size_t nr, size_t kr, size_t sr,
+    size_t k_stride, const uint8_t* k, const int32_t* b, const float* scale,
+    void* packed_weights, size_t extra_bytes,
+    const struct xnn_qs8_qc4w_packing_params* params) {
+  assert(params != nullptr);
+  pack_qs8_qc4uw_gemm_gio_w(
+      g, nc, kc, nr, kr, sr, k_stride, k, b, scale, packed_weights, extra_bytes,
+      params->input_zero_point + 0x80, params->kernel_zero_point);
+}
 void xnn_pack_f32_qs8w_gemm_goi_w(size_t g, size_t nc, size_t kc, size_t nr,
                                   size_t kr, size_t sr, const int8_t* k,
                                   const float* bias, const float* scale,

src/xnnpack/pack.h

@@ -634,6 +634,11 @@ XNN_INTERNAL void xnn_pack_qs8_qc4uw_gemm_gio_w(
     const float* scale, void* packed_weights, size_t extra_bytes,
     const struct xnn_qs8_qc4w_packing_params* params);
 
+XNN_INTERNAL void xnn_pack_qs8_to_qu8_qc4uw_gemm_gio_w(
+    size_t g, size_t nc, size_t kc, size_t nr, size_t kr, size_t sr,
+    size_t k_stride, const uint8_t* kernel, const int32_t* bias,
+    const float* scale, void* packed_weights, size_t extra_bytes,
+    const struct xnn_qs8_qc4w_packing_params* params);
 XNN_INTERNAL void xnn_pack_qs8_qb4w_gemm_gio_w(
     size_t g, size_t nc, size_t kc, size_t nr, size_t kr, size_t sr,
     size_t k_stride, size_t bl, const uint8_t* kernel, const float* bias,