Add MINMAX suffix to remaining micro-kernels
PiperOrigin-RevId: 306067004
diff --git a/BUILD.bazel b/BUILD.bazel
index 690f68d..80c2c5b 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -82,8 +82,8 @@
"src/f32-argmaxpool/4x-scalar-c1.c",
"src/f32-argmaxpool/9p8x-scalar-c1.c",
"src/f32-argmaxpool/9x-scalar-c1.c",
- "src/f32-avgpool/9p8x-scalar-c1.c",
- "src/f32-avgpool/9x-scalar-c1.c",
+ "src/f32-avgpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-avgpool/9x-minmax-scalar-c1.c",
"src/f32-clamp/gen/scalar-x1.c",
"src/f32-clamp/gen/scalar-x2.c",
"src/f32-clamp/gen/scalar-x4.c",
@@ -119,8 +119,8 @@
"src/f32-dwconv/gen/up2x25-minmax-scalar-acc2.c",
"src/f32-dwconv/gen/up2x25-minmax-scalar.c",
"src/f32-gavgpool-spchw/scalar-x1.c",
- "src/f32-gavgpool/7p7x-scalar-c1.c",
- "src/f32-gavgpool/7x-scalar-c1.c",
+ "src/f32-gavgpool/7p7x-minmax-scalar-c1.c",
+ "src/f32-gavgpool/7x-minmax-scalar-c1.c",
"src/f32-gemm/gen-inc/1x4inc-minmax-scalar.c",
"src/f32-gemm/gen-inc/2x4inc-minmax-scalar.c",
"src/f32-gemm/gen-inc/4x4inc-minmax-scalar.c",
@@ -146,9 +146,9 @@
"src/f32-igemm/gen/2x4-minmax-scalar.c",
"src/f32-igemm/gen/4x2-minmax-scalar.c",
"src/f32-igemm/gen/4x4-minmax-scalar.c",
- "src/f32-maxpool/9p8x-scalar-c1.c",
- "src/f32-pavgpool/9p8x-scalar-c1.c",
- "src/f32-pavgpool/9x-scalar-c1.c",
+ "src/f32-maxpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-pavgpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-pavgpool/9x-minmax-scalar-c1.c",
"src/f32-ppmm/gen/2x4-minmax-scalar.c",
"src/f32-ppmm/gen/3x3-minmax-scalar.c",
"src/f32-ppmm/gen/4x2-minmax-scalar.c",
@@ -229,26 +229,26 @@
"src/f32-vbinary/gen/vsubc-minmax-scalar-x1.c",
"src/f32-vbinary/gen/vsubc-minmax-scalar-x2.c",
"src/f32-vbinary/gen/vsubc-minmax-scalar-x4.c",
- "src/f32-vmulcaddc/gen/c1-scalar-2x.c",
- "src/f32-vmulcaddc/gen/c2-scalar-2x.c",
- "src/f32-vmulcaddc/gen/c4-scalar-2x.c",
+ "src/f32-vmulcaddc/gen/c1-minmax-scalar-2x.c",
+ "src/f32-vmulcaddc/gen/c2-minmax-scalar-2x.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-scalar-2x.c",
"src/math/expminus-scalar-lut2048-p1.c",
"src/math/expminus-scalar-lut64-p2.c",
"src/math/expminus-scalar-p5.c",
"src/math/sigmoid-scalar-lut2048-p1-div.c",
"src/math/sigmoid-scalar-lut64-p2-div.c",
"src/math/sigmoid-scalar-p5-div.c",
- "src/q8-avgpool/9p8x-scalar-c1.c",
- "src/q8-avgpool/9x-scalar-c1.c",
+ "src/q8-avgpool/9p8x-minmax-scalar-c1.c",
+ "src/q8-avgpool/9x-minmax-scalar-c1.c",
"src/q8-dwconv/up1x9-minmax-scalar.c",
- "src/q8-gavgpool/7p7x-scalar-c1.c",
- "src/q8-gavgpool/7x-scalar-c1.c",
+ "src/q8-gavgpool/7p7x-minmax-scalar-c1.c",
+ "src/q8-gavgpool/7x-minmax-scalar-c1.c",
"src/q8-gemm/2x2-minmax-scalar.c",
"src/q8-igemm/2x2-minmax-scalar.c",
- "src/q8-vadd/scalar.c",
+ "src/q8-vadd/minmax-scalar.c",
"src/u8-clamp/scalar-x4.c",
"src/u8-lut32norm/scalar.c",
- "src/u8-maxpool/9p8x-scalar-c1.c",
+ "src/u8-maxpool/9p8x-minmax-scalar-c1.c",
"src/u8-rmax/scalar.c",
"src/x32-packx/x2-scalar.c",
"src/x32-packx/x3-scalar.c",
@@ -271,8 +271,8 @@
]
WASM_UKERNELS = [
- "src/f32-avgpool/9p8x-wasm-c1.c",
- "src/f32-avgpool/9x-wasm-c1.c",
+ "src/f32-avgpool/9p8x-minmax-wasm-c1.c",
+ "src/f32-avgpool/9x-minmax-wasm-c1.c",
"src/f32-clamp/gen/wasm-x1.c",
"src/f32-clamp/gen/wasm-x2.c",
"src/f32-clamp/gen/wasm-x4.c",
@@ -300,8 +300,8 @@
"src/f32-dwconv/gen/up2x9-minmax-wasm.c",
"src/f32-dwconv/gen/up2x25-minmax-wasm-acc2.c",
"src/f32-dwconv/gen/up2x25-minmax-wasm.c",
- "src/f32-gavgpool/7p7x-wasm-c1.c",
- "src/f32-gavgpool/7x-wasm-c1.c",
+ "src/f32-gavgpool/7p7x-minmax-wasm-c1.c",
+ "src/f32-gavgpool/7x-minmax-wasm-c1.c",
"src/f32-gemm/gen-inc/1x4inc-minmax-wasm.c",
"src/f32-gemm/gen-inc/2x4inc-minmax-wasm.c",
"src/f32-gemm/gen-inc/4x4inc-minmax-wasm.c",
@@ -324,9 +324,9 @@
"src/f32-igemm/gen/2x4-minmax-wasm.c",
"src/f32-igemm/gen/4x2-minmax-wasm.c",
"src/f32-igemm/gen/4x4-minmax-wasm.c",
- "src/f32-maxpool/9p8x-wasm-c1.c",
- "src/f32-pavgpool/9p8x-wasm-c1.c",
- "src/f32-pavgpool/9x-wasm-c1.c",
+ "src/f32-maxpool/9p8x-minmax-wasm-c1.c",
+ "src/f32-pavgpool/9p8x-minmax-wasm-c1.c",
+ "src/f32-pavgpool/9x-minmax-wasm-c1.c",
"src/f32-vbinary/gen/vadd-minmax-wasm-x1.c",
"src/f32-vbinary/gen/vadd-minmax-wasm-x2.c",
"src/f32-vbinary/gen/vadd-minmax-wasm-x4.c",
@@ -369,17 +369,17 @@
"src/f32-vbinary/gen/vsubc-minmax-wasm-x1.c",
"src/f32-vbinary/gen/vsubc-minmax-wasm-x2.c",
"src/f32-vbinary/gen/vsubc-minmax-wasm-x4.c",
- "src/f32-vmulcaddc/gen/c1-wasm-2x.c",
- "src/f32-vmulcaddc/gen/c2-wasm-2x.c",
- "src/f32-vmulcaddc/gen/c4-wasm-2x.c",
+ "src/f32-vmulcaddc/gen/c1-minmax-wasm-2x.c",
+ "src/f32-vmulcaddc/gen/c2-minmax-wasm-2x.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-wasm-2x.c",
]
PSIMD_FASTMATH_UKERNELS = [
"src/f32-argmaxpool/4x-psimd-c4.c",
"src/f32-argmaxpool/9p8x-psimd-c4.c",
"src/f32-argmaxpool/9x-psimd-c4.c",
- "src/f32-avgpool/9p8x-psimd-c4.c",
- "src/f32-avgpool/9x-psimd-c4.c",
+ "src/f32-avgpool/9p8x-minmax-psimd-c4.c",
+ "src/f32-avgpool/9x-minmax-psimd-c4.c",
"src/f32-clamp/gen/psimd-x4.c",
"src/f32-clamp/gen/psimd-x8.c",
"src/f32-dwconv/gen/up4x25-minmax-psimd-acc2.c",
@@ -394,8 +394,8 @@
"src/f32-dwconv/gen/up8x4-minmax-psimd.c",
"src/f32-dwconv/gen/up8x9-minmax-psimd-acc2.c",
"src/f32-dwconv/gen/up8x9-minmax-psimd.c",
- "src/f32-gavgpool/7p7x-psimd-c4.c",
- "src/f32-gavgpool/7x-psimd-c4.c",
+ "src/f32-gavgpool/7p7x-minmax-psimd-c4.c",
+ "src/f32-gavgpool/7x-minmax-psimd-c4.c",
"src/f32-gemm/gen/1x8-minmax-psimd-loadsplat.c",
"src/f32-gemm/gen/1x8-minmax-psimd-splat.c",
"src/f32-gemm/gen/1x8s4-minmax-psimd.c",
@@ -429,9 +429,9 @@
"src/f32-igemm/gen/6x8-minmax-psimd-loadsplat.c",
"src/f32-igemm/gen/6x8-minmax-psimd-splat.c",
"src/f32-igemm/gen/6x8s4-minmax-psimd.c",
- "src/f32-maxpool/9p8x-psimd-c4.c",
- "src/f32-pavgpool/9p8x-psimd-c4.c",
- "src/f32-pavgpool/9x-psimd-c4.c",
+ "src/f32-maxpool/9p8x-minmax-psimd-c4.c",
+ "src/f32-pavgpool/9p8x-minmax-psimd-c4.c",
+ "src/f32-pavgpool/9x-minmax-psimd-c4.c",
"src/f32-ppmm/gen/4x8-minmax-psimd.c",
"src/f32-prelu/gen/psimd-2x4.c",
"src/f32-prelu/gen/psimd-2x8.c",
@@ -464,8 +464,8 @@
"src/f32-vbinary/gen/vsub-minmax-psimd-x8.c",
"src/f32-vbinary/gen/vsubc-minmax-psimd-x4.c",
"src/f32-vbinary/gen/vsubc-minmax-psimd-x8.c",
- "src/f32-vmulcaddc/gen/c4-psimd-2x.c",
- "src/f32-vmulcaddc/gen/c8-psimd-2x.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-psimd-2x.c",
+ "src/f32-vmulcaddc/gen/c8-minmax-psimd-2x.c",
"src/x32-packx/x4-psimd.c",
"src/x32-pad/x2-psimd.c",
"src/x32-unpool/psimd.c",
@@ -501,8 +501,8 @@
# ISA-specific micro-kernels
NEON_UKERNELS = [
- "src/f32-avgpool/9p8x-neon-c4.c",
- "src/f32-avgpool/9x-neon-c4.c",
+ "src/f32-avgpool/9p8x-minmax-neon-c4.c",
+ "src/f32-avgpool/9x-minmax-neon-c4.c",
"src/f32-clamp/gen/neon-x4.c",
"src/f32-clamp/gen/neon-x8.c",
"src/f32-dwconv/gen/up4x9-minmax-neon.c",
@@ -510,8 +510,8 @@
"src/f32-dwconv/gen/up8x9-minmax-neon.c",
"src/f32-dwconv/gen/up8x9-minmax-neon-acc2.c",
"src/f32-gavgpool-spchw/neon-x4.c",
- "src/f32-gavgpool/7p7x-neon-c4.c",
- "src/f32-gavgpool/7x-neon-c4.c",
+ "src/f32-gavgpool/7p7x-minmax-neon-c4.c",
+ "src/f32-gavgpool/7x-minmax-neon-c4.c",
"src/f32-gemm/gen/1x8-minmax-neon-lane-ld64.c",
"src/f32-gemm/gen/4x2-minmax-neon-lane-ld64.c",
"src/f32-gemm/gen/4x8-minmax-neon-lane-ld128.c",
@@ -563,9 +563,9 @@
"src/f32-igemm/gen/4x8s4-minmax-neon.c",
"src/f32-igemm/gen/6x8s4-minmax-neon.c",
"src/f32-igemm/gen/8x8s4-minmax-neon.c",
- "src/f32-maxpool/9p8x-neon-c4.c",
- "src/f32-pavgpool/9p8x-neon-c4.c",
- "src/f32-pavgpool/9x-neon-c4.c",
+ "src/f32-maxpool/9p8x-minmax-neon-c4.c",
+ "src/f32-pavgpool/9p8x-minmax-neon-c4.c",
+ "src/f32-pavgpool/9x-minmax-neon-c4.c",
"src/f32-ppmm/gen/4x8-minmax-neon.c",
"src/f32-ppmm/gen/8x8-minmax-neon.c",
"src/f32-prelu/gen/neon-2x4.c",
@@ -636,20 +636,20 @@
"src/f32-vbinary/gen/vsub-minmax-neon-x8.c",
"src/f32-vbinary/gen/vsubc-minmax-neon-x4.c",
"src/f32-vbinary/gen/vsubc-minmax-neon-x8.c",
- "src/f32-vmulcaddc/gen/c4-neon-2x.c",
- "src/f32-vmulcaddc/gen/c8-neon-2x.c",
- "src/q8-avgpool/9p8x-neon-c8.c",
- "src/q8-avgpool/9x-neon-c8.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-neon-2x.c",
+ "src/f32-vmulcaddc/gen/c8-minmax-neon-2x.c",
+ "src/q8-avgpool/9p8x-minmax-neon-c8.c",
+ "src/q8-avgpool/9x-minmax-neon-c8.c",
"src/q8-dwconv/up8x9-minmax-neon.c",
- "src/q8-gavgpool/7p7x-neon-c8.c",
- "src/q8-gavgpool/7x-neon-c8.c",
+ "src/q8-gavgpool/7p7x-minmax-neon-c8.c",
+ "src/q8-gavgpool/7x-minmax-neon-c8.c",
"src/q8-gemm/4x8-minmax-neon.c",
"src/q8-gemm/8x8-minmax-neon.c",
"src/q8-igemm/4x8-minmax-neon.c",
"src/q8-igemm/8x8-minmax-neon.c",
- "src/q8-vadd/neon.c",
+ "src/q8-vadd/minmax-neon.c",
"src/u8-clamp/neon-x64.c",
- "src/u8-maxpool/9p8x-neon-c16.c",
+ "src/u8-maxpool/9p8x-minmax-neon-c16.c",
"src/u8-rmax/neon.c",
"src/x32-packx/x4-neon-st4.c",
"src/x32-pad/x2-neon.c",
@@ -790,8 +790,8 @@
"src/f32-sigmoid/gen/neonfma-rr1-lut2048-p1-nr2recps-x16.c",
"src/f32-sigmoid/gen/neonfma-rr1-lut2048-p1-nr2recps-x20.c",
"src/f32-sigmoid/gen/neonfma-rr1-lut2048-p1-nr2recps-x24.c",
- "src/f32-vmulcaddc/gen/c4-neonfma-2x.c",
- "src/f32-vmulcaddc/gen/c8-neonfma-2x.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-neonfma-2x.c",
+ "src/f32-vmulcaddc/gen/c8-minmax-neonfma-2x.c",
"src/math/exp-neonfma-lut64-p2.c",
"src/math/exp-neonfma-p5.c",
"src/math/expminus-neonfma-lut2048-p1.c",
@@ -910,8 +910,8 @@
]
SSE_UKERNELS = [
- "src/f32-avgpool/9p8x-sse-c4.c",
- "src/f32-avgpool/9x-sse-c4.c",
+ "src/f32-avgpool/9p8x-minmax-sse-c4.c",
+ "src/f32-avgpool/9x-minmax-sse-c4.c",
"src/f32-clamp/gen/sse-x4.c",
"src/f32-clamp/gen/sse-x8.c",
"src/f32-dwconv-spchw/3x3p1-sse.c",
@@ -929,8 +929,8 @@
"src/f32-dwconv/gen/up8x9-minmax-sse-acc2.c",
"src/f32-dwconv/gen/up8x9-minmax-sse.c",
"src/f32-gavgpool-spchw/sse-x4.c",
- "src/f32-gavgpool/7p7x-sse-c4.c",
- "src/f32-gavgpool/7x-sse-c4.c",
+ "src/f32-gavgpool/7p7x-minmax-sse-c4.c",
+ "src/f32-gavgpool/7x-minmax-sse-c4.c",
"src/f32-gemm/gen/1x8-minmax-sse-dup.c",
"src/f32-gemm/gen/1x8-minmax-sse-load1.c",
"src/f32-gemm/gen/1x8s4-minmax-sse.c",
@@ -955,9 +955,9 @@
"src/f32-igemm/gen/4x8-minmax-sse-dup.c",
"src/f32-igemm/gen/4x8-minmax-sse-load1.c",
"src/f32-igemm/gen/4x8s4-minmax-sse.c",
- "src/f32-maxpool/9p8x-sse-c4.c",
- "src/f32-pavgpool/9p8x-sse-c4.c",
- "src/f32-pavgpool/9x-sse-c4.c",
+ "src/f32-maxpool/9p8x-minmax-sse-c4.c",
+ "src/f32-pavgpool/9p8x-minmax-sse-c4.c",
+ "src/f32-pavgpool/9x-minmax-sse-c4.c",
"src/f32-ppmm/gen/4x8-minmax-sse.c",
"src/f32-rmax/sse.c",
"src/f32-spmm/gen/4x1-minmax-sse.c",
@@ -990,8 +990,8 @@
"src/f32-vbinary/gen/vsub-minmax-sse-x8.c",
"src/f32-vbinary/gen/vsubc-minmax-sse-x4.c",
"src/f32-vbinary/gen/vsubc-minmax-sse-x8.c",
- "src/f32-vmulcaddc/gen/c4-sse-2x.c",
- "src/f32-vmulcaddc/gen/c8-sse-2x.c",
+ "src/f32-vmulcaddc/gen/c4-minmax-sse-2x.c",
+ "src/f32-vmulcaddc/gen/c8-minmax-sse-2x.c",
"src/x32-packx/x4-sse.c",
]
@@ -1019,17 +1019,17 @@
"src/f32-sigmoid/gen/sse2-p5-div-x16.c",
"src/f32-sigmoid/gen/sse2-p5-div-x20.c",
"src/f32-sigmoid/gen/sse2-p5-div-x24.c",
- "src/q8-avgpool/9p8x-sse2-c8.c",
- "src/q8-avgpool/9x-sse2-c8.c",
+ "src/q8-avgpool/9p8x-minmax-sse2-c8.c",
+ "src/q8-avgpool/9x-minmax-sse2-c8.c",
"src/q8-igemm/4x4c2-minmax-sse2.c",
"src/q8-dwconv/up8x9-minmax-sse2.c",
- "src/q8-gavgpool/7p7x-sse2-c8.c",
- "src/q8-gavgpool/7x-sse2-c8.c",
+ "src/q8-gavgpool/7p7x-minmax-sse2-c8.c",
+ "src/q8-gavgpool/7x-minmax-sse2-c8.c",
"src/q8-gemm/2x4c8-minmax-sse2.c",
"src/q8-gemm/4x4c2-minmax-sse2.c",
- "src/q8-vadd/sse2.c",
+ "src/q8-vadd/minmax-sse2.c",
"src/u8-clamp/sse2-x64.c",
- "src/u8-maxpool/9p8x-sse2-c16.c",
+ "src/u8-maxpool/9p8x-minmax-sse2-c16.c",
"src/u8-rmax/sse2.c",
"src/x32-pad/x2-sse2.c",
"src/x32-zip/x2-sse2.c",
@@ -2491,9 +2491,9 @@
)
xnnpack_unit_test(
- name = "f32_avgpool_test",
+ name = "f32_avgpool_minmax_test",
srcs = [
- "test/f32-avgpool.cc",
+ "test/f32-avgpool-minmax.cc",
"test/avgpool-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_TEST_HDRS,
@@ -2590,9 +2590,9 @@
)
xnnpack_unit_test(
- name = "f32_gavgpool_test",
+ name = "f32_gavgpool_minmax_test",
srcs = [
- "test/f32-gavgpool.cc",
+ "test/f32-gavgpool-minmax.cc",
"test/gavgpool-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_TEST_HDRS,
@@ -2649,18 +2649,18 @@
)
xnnpack_unit_test(
- name = "f32_maxpool_test",
+ name = "f32_maxpool_minmax_test",
srcs = [
- "test/f32-maxpool.cc",
+ "test/f32-maxpool-minmax.cc",
"test/maxpool-microkernel-tester.h",
] + MICROKERNEL_TEST_HDRS,
deps = MICROKERNEL_TEST_DEPS,
)
xnnpack_unit_test(
- name = "f32_pavgpool_test",
+ name = "f32_pavgpool_minmax_test",
srcs = [
- "test/f32-pavgpool.cc",
+ "test/f32-pavgpool-minmax.cc",
"test/avgpool-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_TEST_HDRS,
@@ -2842,9 +2842,9 @@
)
xnnpack_unit_test(
- name = "f32_vmulcaddc_test",
+ name = "f32_vmulcaddc_minmax_test",
srcs = [
- "test/f32-vmulcaddc.cc",
+ "test/f32-vmulcaddc-minmax.cc",
"test/vmulcaddc-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + WEIGHTS_PACK_HDRS + MICROKERNEL_TEST_HDRS,
@@ -2906,9 +2906,9 @@
)
xnnpack_unit_test(
- name = "q8_avgpool_test",
+ name = "q8_avgpool_minmax_test",
srcs = [
- "test/q8-avgpool.cc",
+ "test/q8-avgpool-minmax.cc",
"test/avgpool-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_TEST_HDRS,
@@ -2936,9 +2936,9 @@
)
xnnpack_unit_test(
- name = "q8_gavgpool_test",
+ name = "q8_gavgpool_minmax_test",
srcs = [
- "test/q8-gavgpool.cc",
+ "test/q8-gavgpool-minmax.cc",
"test/gavgpool-microkernel-tester.h",
"src/xnnpack/AlignedAllocator.h",
] + MICROKERNEL_TEST_HDRS,
@@ -2956,9 +2956,9 @@
)
xnnpack_unit_test(
- name = "q8_vadd_test",
+ name = "q8_vadd_minmax_test",
srcs = [
- "test/q8-vadd.cc",
+ "test/q8-vadd-minmax.cc",
"test/vadd-microkernel-tester.h",
] + MICROKERNEL_TEST_HDRS,
deps = MICROKERNEL_TEST_DEPS,
@@ -2983,9 +2983,9 @@
)
xnnpack_unit_test(
- name = "u8_maxpool_test",
+ name = "u8_maxpool_minmax_test",
srcs = [
- "test/u8-maxpool.cc",
+ "test/u8-maxpool-minmax.cc",
"test/maxpool-microkernel-tester.h",
] + MICROKERNEL_TEST_HDRS,
deps = MICROKERNEL_TEST_DEPS,