commit 05f3f6dc940ea45796c009bd09779f597a99151d
author Marat Dukhan <maratek@google.com> Tue Dec 03 15:13:53 2019 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Tue Dec 03 15:14:27 2019 -0800
tree b9c250d88a41c792471a1add0bfa208ac84df404
parent c4f0ff91b796f9023c2766966f7afe36fad14eb1

Subtract ND operator

Binary ND subtraction operator with broadcasting support

PiperOrigin-RevId: 283632285

src/init.c

diff --git a/src/init.c b/src/init.c
index a82716a..5c32056 100644
--- a/src/init.c
+++ b/src/init.c
@@ -230,6 +230,12 @@
       .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vmulc_ukernel__neon_x8,
       .element_tile = 8,
     };
+    xnn_params.f32.vsub = (struct vbinary_parameters) {
+      .op_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsub_ukernel__neon_x8,
+      .opc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsubc_ukernel__neon_x8,
+      .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vrsubc_ukernel__neon_x8,
+      .element_tile = 8,
+    };
     xnn_params.f32.vmulcaddc = (struct vmulcaddc_parameters) {
       .ukernel = (xnn_vmulcaddc_ukernel_function) xnn_f32_vmulcaddc_ukernel_c4__neon_2x,
       .channel_tile = 4,
@@ -510,6 +516,12 @@
       .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vmulc_ukernel__neon_x8,
       .element_tile = 8,
     };
+    xnn_params.f32.vsub = (struct vbinary_parameters) {
+      .op_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsub_ukernel__neon_x8,
+      .opc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsubc_ukernel__neon_x8,
+      .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vrsubc_ukernel__neon_x8,
+      .element_tile = 8,
+    };
     xnn_params.f32.vmulcaddc = (struct vmulcaddc_parameters) {
       .ukernel = (xnn_vmulcaddc_ukernel_function) xnn_f32_vmulcaddc_ukernel_c4__neonfma_2x,
       .channel_tile = 4,
@@ -818,6 +830,12 @@
       .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vmulc_ukernel__sse_x8,
       .element_tile = 8,
     };
+    xnn_params.f32.vsub = (struct vbinary_parameters) {
+      .op_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsub_ukernel__sse_x8,
+      .opc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsubc_ukernel__sse_x8,
+      .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vrsubc_ukernel__sse_x8,
+      .element_tile = 8,
+    };
     xnn_params.f32.vmulcaddc = (struct vmulcaddc_parameters) {
       .ukernel = (xnn_vmulcaddc_ukernel_function) xnn_f32_vmulcaddc_ukernel_c4__sse_2x,
       .channel_tile = 4,
@@ -1024,6 +1042,12 @@
       .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vmulc_ukernel__psimd_x8,
       .element_tile = 8,
     };
+    xnn_params.f32.vsub = (struct vbinary_parameters) {
+      .op_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsub_ukernel__psimd_x8,
+      .opc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsubc_ukernel__psimd_x8,
+      .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vrsubc_ukernel__psimd_x8,
+      .element_tile = 8,
+    };
     xnn_params.f32.vmulcaddc = (struct vmulcaddc_parameters) {
       .ukernel = (xnn_vmulcaddc_ukernel_function) xnn_f32_vmulcaddc_ukernel_c4__psimd_2x,
       .channel_tile = 4,
@@ -1205,6 +1229,12 @@
       .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vmulc_ukernel__scalar_x4,
       .element_tile = 8,
     };
+    xnn_params.f32.vsub = (struct vbinary_parameters) {
+      .op_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsub_ukernel__scalar_x4,
+      .opc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vsubc_ukernel__scalar_x4,
+      .ropc_ukernel = (xnn_vbinary_ukernel_function) xnn_f32_vrsubc_ukernel__scalar_x4,
+      .element_tile = 8,
+    };
     xnn_params.f32.vmulcaddc = (struct vmulcaddc_parameters) {
       .ukernel = (xnn_vmulcaddc_ukernel_function) xnn_f32_vmulcaddc_ukernel_c1__scalar_2x,
       .channel_tile = 1,