Skip to content

Commit

Permalink
Merge branch 'google:master' into img_patch31_qs8_f32_vcvt
Browse files Browse the repository at this point in the history
  • Loading branch information
@KaustubhIMG
KaustubhIMG authored Sep 25, 2024
2 parents 37c2f64 + 90aef2e commit 239a74e
Show file tree
Hide file tree
Showing 1,609 changed files with 50,243 additions and 211,736 deletions.
118 changes: 48 additions & 70 deletions BUILD.bazel
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,6 @@ load(
"xnnpack_min_size_copts",
"xnnpack_slinky_deps",
"xnnpack_slinky_srcs",
"xnnpack_std_cxxopts",
"xnnpack_transitive_source_list",
"xnnpack_visibility",
)
Expand Down Expand Up @@ -87,22 +86,21 @@ SUBGRAPH_SRCS = [
"src/runtime.c",
"src/subgraph.c",
"src/subgraph/abs.c",
"src/subgraph/add2.c",
"src/subgraph/argmax-pooling-2d.c",
"src/subgraph/average-pooling-2d.c",
"src/subgraph/bankers-rounding.c",
"src/subgraph/batch-matrix-multiply.c",
"src/subgraph/binary.c",
"src/subgraph/ceiling.c",
"src/subgraph/clamp.c",
"src/subgraph/concatenate.c",
"src/subgraph/convert.c",
"src/subgraph/convolution-2d.c",
"src/subgraph/copy.c",
"src/subgraph/copysign.c",
"src/subgraph/deconvolution-2d.c",
"src/subgraph/deprecated.c",
"src/subgraph/depth-to-space-2d.c",
"src/subgraph/depthwise-convolution-2d.c",
"src/subgraph/divide.c",
"src/subgraph/elu.c",
"src/subgraph/even-split.c",
"src/subgraph/exp.c",
Expand All @@ -116,9 +114,6 @@ SUBGRAPH_SRCS = [
"src/subgraph/leaky-relu.c",
"src/subgraph/log.c",
"src/subgraph/max-pooling-2d.c",
"src/subgraph/maximum2.c",
"src/subgraph/minimum2.c",
"src/subgraph/multiply2.c",
"src/subgraph/negate.c",
"src/subgraph/prelu.c",
"src/subgraph/reciprocal-square-root.c",
Expand All @@ -130,13 +125,11 @@ SUBGRAPH_SRCS = [
"src/subgraph/space-to-depth-2d.c",
"src/subgraph/square-root.c",
"src/subgraph/square.c",
"src/subgraph/squared-difference.c",
"src/subgraph/static-constant-pad.c",
"src/subgraph/static-mean.c",
"src/subgraph/static-resize-bilinear-2d.c",
"src/subgraph/static-slice.c",
"src/subgraph/static-transpose.c",
"src/subgraph/subtract.c",
"src/subgraph/tanh.c",
"src/subgraph/unpooling-2d.c",
"src/subgraph/validation.c",
Expand All @@ -161,26 +154,26 @@ MICROKERNEL_DEFS = [
"src/f16-maxpool/f16-maxpool-minmax.h",
"src/f16-pavgpool/f16-pavgpool-minmax.h",
"src/f16-vabs/f16-vabs.h",
"src/f16-vbinary/f16-vadd-minmax.h",
"src/f16-vbinary/f16-vaddc-minmax.h",
"src/f16-vbinary/f16-vadd.h",
"src/f16-vbinary/f16-vaddc.h",
"src/f16-vbinary/f16-vcmul.h",
"src/f16-vbinary/f16-vdiv-minmax.h",
"src/f16-vbinary/f16-vdivc-minmax.h",
"src/f16-vbinary/f16-vdiv.h",
"src/f16-vbinary/f16-vdivc.h",
"src/f16-vbinary/f16-vmax.h",
"src/f16-vbinary/f16-vmaxc.h",
"src/f16-vbinary/f16-vmin.h",
"src/f16-vbinary/f16-vminc.h",
"src/f16-vbinary/f16-vmul-minmax.h",
"src/f16-vbinary/f16-vmulc-minmax.h",
"src/f16-vbinary/f16-vmul.h",
"src/f16-vbinary/f16-vmulc.h",
"src/f16-vbinary/f16-vprelu.h",
"src/f16-vbinary/f16-vpreluc.h",
"src/f16-vbinary/f16-vrdivc-minmax.h",
"src/f16-vbinary/f16-vrdivc.h",
"src/f16-vbinary/f16-vrpreluc.h",
"src/f16-vbinary/f16-vrsubc-minmax.h",
"src/f16-vbinary/f16-vrsubc.h",
"src/f16-vbinary/f16-vsqrdiff.h",
"src/f16-vbinary/f16-vsqrdiffc.h",
"src/f16-vbinary/f16-vsub-minmax.h",
"src/f16-vbinary/f16-vsubc-minmax.h",
"src/f16-vbinary/f16-vsub.h",
"src/f16-vbinary/f16-vsubc.h",
"src/f16-vclamp/f16-vclamp.h",
"src/f16-velu/f16-velu.h",
"src/f16-vhswish/f16-vhswish.h",
Expand All @@ -199,38 +192,28 @@ MICROKERNEL_DEFS = [
"src/f32-maxpool/f32-maxpool-minmax.h",
"src/f32-pavgpool/f32-pavgpool-minmax.h",
"src/f32-vabs/f32-vabs.h",
"src/f32-vbinary/f32-vadd-minmax.h",
"src/f32-vbinary/f32-vadd.h",
"src/f32-vbinary/f32-vaddc-minmax.h",
"src/f32-vbinary/f32-vaddc.h",
"src/f32-vbinary/f32-vcmul.h",
"src/f32-vbinary/f32-vcopysign.h",
"src/f32-vbinary/f32-vcopysignc.h",
"src/f32-vbinary/f32-vdiv-minmax.h",
"src/f32-vbinary/f32-vdiv.h",
"src/f32-vbinary/f32-vdivc-minmax.h",
"src/f32-vbinary/f32-vdivc.h",
"src/f32-vbinary/f32-vmax.h",
"src/f32-vbinary/f32-vmaxc.h",
"src/f32-vbinary/f32-vmin.h",
"src/f32-vbinary/f32-vminc.h",
"src/f32-vbinary/f32-vmul-minmax.h",
"src/f32-vbinary/f32-vmul.h",
"src/f32-vbinary/f32-vmulc-minmax.h",
"src/f32-vbinary/f32-vmulc.h",
"src/f32-vbinary/f32-vprelu.h",
"src/f32-vbinary/f32-vpreluc.h",
"src/f32-vbinary/f32-vrcopysignc.h",
"src/f32-vbinary/f32-vrdivc-minmax.h",
"src/f32-vbinary/f32-vrdivc.h",
"src/f32-vbinary/f32-vrpreluc.h",
"src/f32-vbinary/f32-vrsubc-minmax.h",
"src/f32-vbinary/f32-vrsubc.h",
"src/f32-vbinary/f32-vsqrdiff.h",
"src/f32-vbinary/f32-vsqrdiffc.h",
"src/f32-vbinary/f32-vsub-minmax.h",
"src/f32-vbinary/f32-vsub.h",
"src/f32-vbinary/f32-vsubc-minmax.h",
"src/f32-vbinary/f32-vsubc.h",
"src/f32-vclamp/f32-vclamp.h",
"src/f32-velu/f32-velu.h",
Expand Down Expand Up @@ -274,7 +257,6 @@ MICROKERNEL_DEFS = [
"src/s32-vmul/s32-vmulc.h",
"src/u8-maxpool/u8-maxpool-minmax.h",
"src/u8-vclamp/u8-vclamp.h",
"src/u64-u32-vsqrtshift/u64-u32-vsqrtshift.h",
"src/x8-packq/x8-packq.h",
"src/x8-packw/x8-packw.h",
"src/x16-packw/x16-packw.h",
Expand Down Expand Up @@ -303,16 +285,30 @@ MICROKERNEL_DEFS = [
"src/xx-transposev/xx-transposev.h",
"src/xx-fill/xx-fill.h",
"src/xx-pad/xx-pad.h",
"src/f16-dwconv/f16-dwconv-minmax-unipass.h",
"src/f32-dwconv/f32-dwconv-minmax-unipass.h",
"src/f32-dwconv/f32-dwconv-unipass.h",
"src/qs8-dwconv/qs8-dwconv-minmax-unipass-fp32.h",
"src/qs8-dwconv/qs8-dwconv-minmax-unipass-rndnu.h",
"src/qs8-qc8w-dwconv/qs8-qc8w-dwconv-minmax-unipass-fp32.h",
"src/qu8-dwconv/qu8-dwconv-minmax-unipass-fp32.h",
"src/qu8-dwconv/qu8-dwconv-minmax-unipass-rndnu.h",
"src/f16-dwconv/f16-dwconv-minmax-multipass.h",
"src/f32-dwconv/f32-dwconv-minmax-multipass.h",
"src/f32-dwconv/f32-dwconv-multipass.h",
"src/qs8-dwconv/qs8-dwconv-minmax-multipass-fp32.h",
"src/qs8-dwconv/qs8-dwconv-minmax-multipass-rndnu.h",
"src/qs8-qc8w-dwconv/qs8-qc8w-dwconv-minmax-multipass-fp32.h",
"src/qu8-dwconv/qu8-dwconv-minmax-multipass-fp32.h",
"src/qu8-dwconv/qu8-dwconv-minmax-multipass-rndnu.h",
]

MICROKERNEL_HDRS = [
"src/xnnpack/argmaxpool.h",
"src/xnnpack/avgpool.h",
"src/xnnpack/conv.h",
"src/xnnpack/dwconv.h",
"src/xnnpack/fft.h",
"src/xnnpack/fill.h",
"src/xnnpack/filterbank.h",
"src/xnnpack/gavgpool.h",
"src/xnnpack/gemm.h",
"src/xnnpack/ibilinear.h",
Expand All @@ -332,7 +328,6 @@ MICROKERNEL_HDRS = [
"src/xnnpack/raddextexp.h",
"src/xnnpack/raddstoreexpminusmax.h",
"src/xnnpack/reduce.h",
"src/xnnpack/rmaxabs.h",
"src/xnnpack/spmm.h",
"src/xnnpack/transpose.h",
"src/xnnpack/unpool.h",
Expand All @@ -341,13 +336,10 @@ MICROKERNEL_HDRS = [
"src/xnnpack/vhswish.h",
"src/xnnpack/vlog.h",
"src/xnnpack/vlrelu.h",
"src/xnnpack/vlshift.h",
"src/xnnpack/vmulcaddc.h",
"src/xnnpack/vscaleexpminusmax.h",
"src/xnnpack/vscaleextexp.h",
"src/xnnpack/vsquareabs.h",
"src/xnnpack/vunary.h",
"src/xnnpack/window.h",
"src/xnnpack/zerob.h",
"src/xnnpack/zip.h",
] + MICROKERNEL_DEFS
Expand Down Expand Up @@ -554,6 +546,7 @@ xnnpack_cc_library(
":microparams_init",
":packing",
":prod_microkernels",
":xnnpack_h",
"@FP16",
] + select({
":cpuinfo_enabled": ["@cpuinfo"],
Expand Down Expand Up @@ -612,37 +605,48 @@ xnnpack_cc_library(
":microparams",
":unaligned",
":xnnpack_h",
"@FP16",
],
)

xnnpack_cc_library(
name = "node_type",
hdrs = ["src/xnnpack/node-type.h"],
hdrs = [
"src/xnnpack/node-type.h",
"src/xnnpack/node-type-defs.h",
],
deps = [
":common",
],
)

xnnpack_cc_library(
name = "allocation_type",
hdrs = ["src/xnnpack/allocation-type.h"],
hdrs = [
"src/xnnpack/allocation-type.h",
"src/xnnpack/allocation-type-defs.h",
],
deps = [
":common",
],
)

xnnpack_cc_library(
name = "operator_type",
hdrs = ["src/xnnpack/operator-type.h"],
hdrs = [
"src/xnnpack/operator-type.h",
"src/xnnpack/operator-type-defs.h",
],
deps = [
":common",
],
)

xnnpack_cc_library(
name = "microkernel_type",
hdrs = ["src/xnnpack/microkernel-type.h"],
hdrs = [
"src/xnnpack/microkernel-type.h",
"src/xnnpack/microkernel-type-defs.h",
],
deps = [
":common",
],
Expand Down Expand Up @@ -789,18 +793,6 @@ filegroup(
"fi",
compatible_with = [],
),
genrule(
name = arch + "_non_prod_microkernel_srcs",
srcs = non_prod_c_srcs_for_arch(arch),
outs = [arch + "_non_prod_microkernels.c"],
cmd = "if [ -z \"$(SRCS)\" ]; then " +
" echo \"\" > $@; " +
"else " +
" cat $(SRCS) | grep -E '^#include ' | sort -u > $@; " +
" cat $(SRCS) | grep -v -E '^#include ' >> $@; " +
"fi",
compatible_with = [],
),
xnnpack_cc_library_for_arch(
name = arch + "_prod_microkernels",
srcs = prod_asm_srcs_for_arch(arch) + [":" + arch + "_prod_microkernel_srcs"],
Expand All @@ -812,7 +804,7 @@ filegroup(
),
xnnpack_cc_library_for_arch(
name = arch + "_all_microkernels",
srcs = non_prod_asm_srcs_for_arch(arch) + [":" + arch + "_non_prod_microkernel_srcs"],
srcs = non_prod_asm_srcs_for_arch(arch) + non_prod_c_srcs_for_arch(arch),
arch = arch,
compatible_with = [],
defines = xnnpack_configurable_defines(),
Expand Down Expand Up @@ -1049,9 +1041,9 @@ xnnpack_cc_library(
":logging",
":math",
":operator_h",
":operator_type",
":params",
":xnnpack_h",
"@FP16",
],
)

Expand Down Expand Up @@ -1246,21 +1238,6 @@ xnnpack_cc_library(
],
)

############################# End-to-end benchmarks ############################

# Helper library for benchmarks to depend on.
xnnpack_cc_library(
name = "models_h",
hdrs = ["src/xnnpack/models.h"],
copts = xnnpack_std_cxxopts(),
deps = [
":XNNPACK",
":aligned_allocator",
":common",
"@FP16",
],
)

############################# Build configurations #############################

# Enables usage of ARM FP16 (FP16 arithmetics) scalar kernels.
Expand Down Expand Up @@ -1797,6 +1774,7 @@ alias(
":xnn_enable_avxvnniint8_explicit_true": ":xnn_enable_avxvnniint8_explicit_true",
":xnn_enable_avxvnniint8_explicit_false": ":xnn_enable_avxvnniint8_explicit_true",
"//build_config:ios_x86_64": ":xnn_enable_avxvnniint8_explicit_true",
"//build_config:ios": ":xnn_enable_avxvnniint8_explicit_true",
"//conditions:default": ":avxvnniint8_enabled_by_default",
}),
)
Expand Down
Loading

0 comments on commit 239a74e

Please sign in to comment.