X32-PACKW Neon microkernels renamed to include ld4lane and ld2lane

- x8 and x12 add ld4lane suffix - x2 add ld2lane suffix PiperOrigin-RevId: 516538969
L1onKing · Mar 14, 2023 · 9ae8348 · 9ae8348
1 parent daad9f6
commit 9ae8348
Show file tree

Hide file tree

Showing 13 changed files with 132 additions and 132 deletions.
diff --git a/bench/f32-bgemm.cc b/bench/f32-bgemm.cc
diff --git a/bench/x32-packw.cc b/bench/x32-packw.cc
@@ -81,28 +81,28 @@ static void x32_packw(benchmark::State& state,
 
 
 #if XNN_ARCH_ARM || XNN_ARCH_ARM64
-  static void x32_packw_x2__neon(benchmark::State& state, const char* net) {
+  static void x32_packw_x2__neon_ld2lane(benchmark::State& state, const char* net) {
     x32_packw(state,
-      xnn_x32_packw_gemm_goi_ukernel_x2__neon,
+      xnn_x32_packw_gemm_goi_ukernel_x2__neon_ld2lane,
       /*nr=*/2, /*kr=*/1, /*sr=*/1,
       benchmark::utils::CheckNEON);
   }
-  static void x32_packw_x8__neon(benchmark::State& state, const char* net) {
+  static void x32_packw_x8__neon_ld4lane(benchmark::State& state, const char* net) {
     x32_packw(state,
-      xnn_x32_packw_gemm_goi_ukernel_x8__neon,
+      xnn_x32_packw_gemm_goi_ukernel_x8__neon_ld4lane,
       /*nr=*/8, /*kr=*/1, /*sr=*/1,
       benchmark::utils::CheckNEON);
   }
-  static void x32_packw_x12__neon(benchmark::State& state, const char* net) {
+  static void x32_packw_x12__neon_ld4lane(benchmark::State& state, const char* net) {
     x32_packw(state,
-      xnn_x32_packw_gemm_goi_ukernel_x12__neon,
+      xnn_x32_packw_gemm_goi_ukernel_x12__neon_ld4lane,
       /*nr=*/12, /*kr=*/1, /*sr=*/1,
       benchmark::utils::CheckNEON);
   }
 
-  BENCHMARK_BGEMM(x32_packw_x2__neon)
-  BENCHMARK_BGEMM(x32_packw_x8__neon)
-  BENCHMARK_BGEMM(x32_packw_x12__neon)
+  BENCHMARK_BGEMM(x32_packw_x2__neon_ld2lane)
+  BENCHMARK_BGEMM(x32_packw_x8__neon_ld4lane)
+  BENCHMARK_BGEMM(x32_packw_x12__neon_ld4lane)
 #endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
 
 static void x32_packw_x2__scalar_float(benchmark::State& state, const char* net) {

diff --git a/cmake/microkernels.cmake b/cmake/microkernels.cmake
@@ -3101,9 +3101,9 @@ SET(ALL_NEON_MICROKERNEL_SRCS
   src/x16-transposec/gen/x16-transposec-8x8-reuse-multi-zip-neon.c
   src/x16-transposec/gen/x16-transposec-8x8-reuse-switch-zip-neon.c
   src/x24-transposec/x24-transposec-2x2-neon-tbl64.c
-  src/x32-packw/gen/x32-packw-x2-neon.c
-  src/x32-packw/gen/x32-packw-x8-neon.c
-  src/x32-packw/gen/x32-packw-x12-neon.c
+  src/x32-packw/gen/x32-packw-x2-neon-ld2lane.c
+  src/x32-packw/gen/x32-packw-x8-neon-ld4lane.c
+  src/x32-packw/gen/x32-packw-x12-neon-ld4lane.c
   src/x32-packx/x32-packx-x4-neon-st4.c
   src/x32-transposec/gen/x32-transposec-2x2-multi-dec-zip-neon.c
   src/x32-transposec/gen/x32-transposec-2x2-multi-mov-zip-neon.c
@@ -5119,6 +5119,8 @@ SET(ALL_SCALAR_MICROKERNEL_SRCS
   src/x8-lut/gen/x8-lut-scalar-x4.c
   src/x8-lut/gen/x8-lut-scalar-x8.c
   src/x8-lut/gen/x8-lut-scalar-x16.c
+  src/x8-packw/gen/x8-packw-x2-scalar-int.c
+  src/x8-packw/gen/x8-packw-x4-scalar-int.c
   src/x8-transposec/gen/x8-transposec-1x2-scalar-int.c
   src/x8-transposec/gen/x8-transposec-1x4-scalar-int.c
   src/x8-transposec/gen/x8-transposec-2x1-scalar-int.c
@@ -5131,8 +5133,6 @@ SET(ALL_SCALAR_MICROKERNEL_SRCS
   src/x8-zip/x8-zip-x3-scalar.c
   src/x8-zip/x8-zip-x4-scalar.c
   src/x8-zip/x8-zip-xm-scalar.c
-  src/x8-packw/gen/x8-packw-x2-scalar-int.c
-  src/x8-packw/gen/x8-packw-x4-scalar-int.c
   src/x16-packw/gen/x16-packw-x8-scalar-int.c
   src/x16-packw/gen/x16-packw-x16-scalar-int.c
   src/x16-transposec/gen/x16-transposec-1x2-scalar-int.c

diff --git a/microkernels.bzl b/microkernels.bzl
@@ -3108,9 +3108,9 @@ ALL_NEON_MICROKERNEL_SRCS = [
     "src/x16-transposec/gen/x16-transposec-8x8-reuse-multi-zip-neon.c",
     "src/x16-transposec/gen/x16-transposec-8x8-reuse-switch-zip-neon.c",
     "src/x24-transposec/x24-transposec-2x2-neon-tbl64.c",
-    "src/x32-packw/gen/x32-packw-x2-neon.c",
-    "src/x32-packw/gen/x32-packw-x8-neon.c",
-    "src/x32-packw/gen/x32-packw-x12-neon.c",
+    "src/x32-packw/gen/x32-packw-x2-neon-ld2lane.c",
+    "src/x32-packw/gen/x32-packw-x8-neon-ld4lane.c",
+    "src/x32-packw/gen/x32-packw-x12-neon-ld4lane.c",
     "src/x32-packx/x32-packx-x4-neon-st4.c",
     "src/x32-transposec/gen/x32-transposec-2x2-multi-dec-zip-neon.c",
     "src/x32-transposec/gen/x32-transposec-2x2-multi-mov-zip-neon.c",
@@ -5138,6 +5138,8 @@ ALL_SCALAR_MICROKERNEL_SRCS = [
     "src/x8-lut/gen/x8-lut-scalar-x4.c",
     "src/x8-lut/gen/x8-lut-scalar-x8.c",
     "src/x8-lut/gen/x8-lut-scalar-x16.c",
+    "src/x8-packw/gen/x8-packw-x2-scalar-int.c",
+    "src/x8-packw/gen/x8-packw-x4-scalar-int.c",
     "src/x8-transposec/gen/x8-transposec-1x2-scalar-int.c",
     "src/x8-transposec/gen/x8-transposec-1x4-scalar-int.c",
     "src/x8-transposec/gen/x8-transposec-2x1-scalar-int.c",
@@ -5150,8 +5152,6 @@ ALL_SCALAR_MICROKERNEL_SRCS = [
     "src/x8-zip/x8-zip-x3-scalar.c",
     "src/x8-zip/x8-zip-x4-scalar.c",
     "src/x8-zip/x8-zip-xm-scalar.c",
-    "src/x8-packw/gen/x8-packw-x2-scalar-int.c",
-    "src/x8-packw/gen/x8-packw-x4-scalar-int.c",
     "src/x16-packw/gen/x16-packw-x8-scalar-int.c",
     "src/x16-packw/gen/x16-packw-x16-scalar-int.c",
     "src/x16-transposec/gen/x16-transposec-1x2-scalar-int.c",

diff --git a/scripts/generate-x32-packw.sh b/scripts/generate-x32-packw.sh
@@ -13,11 +13,11 @@ tools/xngen src/x32-packw/scalar.c.in -D NR=4 -D KUNROLL=4 -D TYPE=float    -o s
 
 ################################### ARM NEON ##################################
 ### NR multiple of 4
-tools/xngen src/x32-packw/neon.c.in -D NR=8  -D KUNROLL=4 -o src/x32-packw/gen/x32-packw-x8-neon.c &
-tools/xngen src/x32-packw/neon.c.in -D NR=12 -D KUNROLL=4 -o src/x32-packw/gen/x32-packw-x12-neon.c &
+tools/xngen src/x32-packw/neon.c.in -D NR=8  -D KUNROLL=4 -o src/x32-packw/gen/x32-packw-x8-neon-ld4lane.c &
+tools/xngen src/x32-packw/neon.c.in -D NR=12 -D KUNROLL=4 -o src/x32-packw/gen/x32-packw-x12-neon-ld4lane.c &
 
 ### NR2 micro-kernels
-tools/xngen src/x32-packw/NR2-neon.c.in -D NR=2 -D KUNROLL=2 -o src/x32-packw/gen/x32-packw-x2-neon.c &
+tools/xngen src/x32-packw/NR2-neon.c.in -D NR=2 -D KUNROLL=2 -o src/x32-packw/gen/x32-packw-x2-neon-ld2lane.c &
 
 ################################## Unit tests #################################
 tools/generate-packw-test.py --spec test/x32-packw.yaml --output test/x32-packw.cc &

diff --git a/src/x32-packw/NR2-neon.c.in b/src/x32-packw/NR2-neon.c.in
@@ -16,7 +16,7 @@ $assert KUNROLL >= 2
 #include <xnnpack/packw.h>
 
 
-void xnn_x32_packw_gemm_goi_ukernel_x${NR}__neon(
+void xnn_x32_packw_gemm_goi_ukernel_x${NR}__neon_ld2lane(
   size_t g,
   size_t nc,
   size_t kc,

diff --git a/src/x32-packw/gen/x32-packw-x12-neon.c → ...32-packw/gen/x32-packw-x12-neon-ld4lane.c b/src/x32-packw/gen/x32-packw-x12-neon.c → ...32-packw/gen/x32-packw-x12-neon-ld4lane.c
@@ -17,7 +17,7 @@
 #include <xnnpack/packw.h>
 
 
-void xnn_x32_packw_gemm_goi_ukernel_x12__neon(
+void xnn_x32_packw_gemm_goi_ukernel_x12__neon_ld4lane(
   size_t g,
   size_t nc,
   size_t kc,

diff --git a/src/x32-packw/gen/x32-packw-x2-neon.c → ...x32-packw/gen/x32-packw-x2-neon-ld2lane.c b/src/x32-packw/gen/x32-packw-x2-neon.c → ...x32-packw/gen/x32-packw-x2-neon-ld2lane.c
@@ -18,7 +18,7 @@
 #include <xnnpack/packw.h>
 
 
-void xnn_x32_packw_gemm_goi_ukernel_x2__neon(
+void xnn_x32_packw_gemm_goi_ukernel_x2__neon_ld2lane(
   size_t g,
   size_t nc,
   size_t kc,

diff --git a/src/x32-packw/gen/x32-packw-x8-neon.c → ...x32-packw/gen/x32-packw-x8-neon-ld4lane.c b/src/x32-packw/gen/x32-packw-x8-neon.c → ...x32-packw/gen/x32-packw-x8-neon-ld4lane.c
@@ -17,7 +17,7 @@
 #include <xnnpack/packw.h>
 
 
-void xnn_x32_packw_gemm_goi_ukernel_x8__neon(
+void xnn_x32_packw_gemm_goi_ukernel_x8__neon_ld4lane(
   size_t g,
   size_t nc,
   size_t kc,

diff --git a/src/x32-packw/neon.c.in b/src/x32-packw/neon.c.in
@@ -19,7 +19,7 @@ $ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
 #include <xnnpack/packw.h>
 
 
-void xnn_x32_packw_gemm_goi_ukernel_x${NR}__neon(
+void xnn_x32_packw_gemm_goi_ukernel_x${NR}__neon_ld4lane(
   size_t g,
   size_t nc,
   size_t kc,

diff --git a/src/xnnpack/packw.h b/src/xnnpack/packw.h
@@ -73,9 +73,9 @@ DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x4__s
 DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x2__scalar_float)
 DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x4__scalar_float)
 
-DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x2__neon)
-DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x8__neon)
-DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x12__neon)
+DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x2__neon_ld2lane)
+DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x8__neon_ld4lane)
+DECLARE_X32_PACKW_GEMM_GOI_UKERNEL_FUNCTION(xnn_x32_packw_gemm_goi_ukernel_x12__neon_ld4lane)
 
 
 #ifdef __cplusplus