[PATCH] lib/meson: refactor GPGPU assembly build process
Kamil Konieczny
kamil.konieczny at linux.intel.com
Fri May 30 11:10:19 UTC 2025
Hi Andrzej,
On 2025-05-29 at 14:32:23 +0200, Andrzej Hajda wrote:
> GPGPU inline assembly was build into one big code file
> inside libigt library. This approach has few drawbacks:
Add newline here.
> 1. Only files from the library can contain assembly.
> 2. Change of one assembly chunk results in re-build of
> all chunks.
> 3. Most shaders lands in lib/gpgpu_shader.c.
Add newline.
> With this change assembly codes are generated and stored
> in files relative to files containing asms. For example
> lib/gpgpu_shader.c will have codes in
> lib/gpgpu_shader.c.gen.iga64_codes.c.
Remove last dot from filename, add newline.
> It will allow to move asms specific for particular test
> into the test files, instead of putting them to lib/gpgpu_shader.c.
What about a little more explanation here, like:
It will allow to move asms specific for particular test
into the test files, instead of putting all of them into
lib/gpgpu_shader.c. In result, GPGPU fill assemblies will
be sparated into new one.
Up to you.
>
> Signed-off-by: Andrzej Hajda <andrzej.hajda at intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny at linux.intel.com>
> ---
> This is just upstreamed version of internal patch.
> ---
> lib/gpgpu_fill.c.gen.iga64_codes.c | 80 ++++++++++++++++++++++
> ...ed_codes.c => gpgpu_shader.c.gen.iga64_codes.c} | 76 +-------------------
> lib/meson.build | 40 ++++++-----
> scripts/generate_iga64_codes | 14 ++--
> 4 files changed, 112 insertions(+), 98 deletions(-)
>
> diff --git a/lib/gpgpu_fill.c.gen.iga64_codes.c b/lib/gpgpu_fill.c.gen.iga64_codes.c
> new file mode 100644
> index 000000000000..b223425d049a
> --- /dev/null
> +++ b/lib/gpgpu_fill.c.gen.iga64_codes.c
> @@ -0,0 +1,80 @@
> +// SPDX-License-Identifier: MIT
> +// Generated using Intel Graphics Assembler 1.1.0-int
> +
> +#include "gpgpu_shader.h"
> +
> +#define MD5_SUM_IGA64_ASMS 8304c8bb11385470bfd6ac32e1c36840
> +
> +struct iga64_template const iga64_code_gpgpu_fill[] = {
> + { .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> + 0x00080061, 0x01050000, 0x00000104, 0x00000000,
> + 0x00000069, 0x02058220, 0x02000014, 0x00000004,
> + 0x00000061, 0x02150220, 0x00000064, 0x00000000,
> + 0x000c0061, 0x04054220, 0x00000000, 0x00000000,
> + 0x00101c61, 0x05050220, 0x00000104, 0x00000000,
> + 0x00041b61, 0x04550220, 0x00220205, 0x00000000,
> + 0x00000061, 0x04754550, 0x00000000, 0x000f000f,
> + 0x00132031, 0x00000000, 0xd00e0494, 0x04000000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> + }},
> + { .gen_ver = 1270, .size = 52, .code = (const uint32_t []) {
> + 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> + 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> + 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> + 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> + 0x00041c61, 0x05050220, 0x00000104, 0x00000000,
> + 0x00011b61, 0x04050220, 0x00220205, 0x00000000,
> + 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> + 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> + 0x80001901, 0x00010000, 0x00000000, 0x00000000,
> + 0x00044031, 0x00000000, 0xc0000414, 0x02a00000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> + }},
> + { .gen_ver = 1260, .size = 48, .code = (const uint32_t []) {
> + 0x00080061, 0x01050000, 0x00000104, 0x00000000,
> + 0x00000069, 0x02058220, 0x02000014, 0x00000004,
> + 0x00000061, 0x02150220, 0x00000064, 0x00000000,
> + 0x000c0061, 0x04054220, 0x00000000, 0x00000000,
> + 0x00101c61, 0x05050220, 0x00000104, 0x00000000,
> + 0x00041b61, 0x04050220, 0x00220205, 0x00000000,
> + 0x00000061, 0x04254220, 0x00000000, 0x0000000f,
> + 0x00000061, 0x04450220, 0x00000054, 0x00000000,
> + 0x00132031, 0x00000000, 0xc0000414, 0x02a00000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> + }},
> + { .gen_ver = 1250, .size = 52, .code = (const uint32_t []) {
> + 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> + 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> + 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> + 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> + 0x00041c61, 0x05050220, 0x00000104, 0x00000000,
> + 0x00011b61, 0x04050220, 0x00220205, 0x00000000,
> + 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> + 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> + 0x80001901, 0x00010000, 0x00000000, 0x00000000,
> + 0x00044031, 0x00000000, 0xc0000414, 0x02a00000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> + }},
> + { .gen_ver = 0, .size = 48, .code = (const uint32_t []) {
> + 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> + 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> + 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> + 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> + 0x00040461, 0x05050220, 0x00000104, 0x00000000,
> + 0x00010361, 0x04050220, 0x00220205, 0x00000000,
> + 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> + 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> + 0x00049031, 0x00000000, 0xc0000414, 0x02a00000,
> + 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> + 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> + 0x80000101, 0x00010000, 0x00000000, 0x00000000,
> + }}
> +};
> diff --git a/lib/iga64_generated_codes.c b/lib/gpgpu_shader.c.gen.iga64_codes.c
> similarity index 89%
> rename from lib/iga64_generated_codes.c
> rename to lib/gpgpu_shader.c.gen.iga64_codes.c
> index 551dead2be17..fffe242fe138 100644
> --- a/lib/iga64_generated_codes.c
> +++ b/lib/gpgpu_shader.c.gen.iga64_codes.c
> @@ -3,81 +3,7 @@
>
> #include "gpgpu_shader.h"
>
> -#define MD5_SUM_IGA64_ASMS 22a5a4671242c3257209f4051f1c0c7e
> -
> -struct iga64_template const iga64_code_gpgpu_fill[] = {
> - { .gen_ver = 2000, .size = 44, .code = (const uint32_t []) {
> - 0x00080061, 0x01050000, 0x00000104, 0x00000000,
> - 0x00000069, 0x02058220, 0x02000014, 0x00000004,
> - 0x00000061, 0x02150220, 0x00000064, 0x00000000,
> - 0x000c0061, 0x04054220, 0x00000000, 0x00000000,
> - 0x00101c61, 0x05050220, 0x00000104, 0x00000000,
> - 0x00041b61, 0x04550220, 0x00220205, 0x00000000,
> - 0x00000061, 0x04754550, 0x00000000, 0x000f000f,
> - 0x00132031, 0x00000000, 0xd00e0494, 0x04000000,
> - 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> - 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> - 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> - }},
> - { .gen_ver = 1270, .size = 52, .code = (const uint32_t []) {
> - 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> - 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> - 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> - 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> - 0x00041c61, 0x05050220, 0x00000104, 0x00000000,
> - 0x00011b61, 0x04050220, 0x00220205, 0x00000000,
> - 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> - 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> - 0x80001901, 0x00010000, 0x00000000, 0x00000000,
> - 0x00044031, 0x00000000, 0xc0000414, 0x02a00000,
> - 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> - 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> - 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> - }},
> - { .gen_ver = 1260, .size = 48, .code = (const uint32_t []) {
> - 0x00080061, 0x01050000, 0x00000104, 0x00000000,
> - 0x00000069, 0x02058220, 0x02000014, 0x00000004,
> - 0x00000061, 0x02150220, 0x00000064, 0x00000000,
> - 0x000c0061, 0x04054220, 0x00000000, 0x00000000,
> - 0x00101c61, 0x05050220, 0x00000104, 0x00000000,
> - 0x00041b61, 0x04050220, 0x00220205, 0x00000000,
> - 0x00000061, 0x04254220, 0x00000000, 0x0000000f,
> - 0x00000061, 0x04450220, 0x00000054, 0x00000000,
> - 0x00132031, 0x00000000, 0xc0000414, 0x02a00000,
> - 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> - 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> - 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> - }},
> - { .gen_ver = 1250, .size = 52, .code = (const uint32_t []) {
> - 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> - 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> - 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> - 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> - 0x00041c61, 0x05050220, 0x00000104, 0x00000000,
> - 0x00011b61, 0x04050220, 0x00220205, 0x00000000,
> - 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> - 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> - 0x80001901, 0x00010000, 0x00000000, 0x00000000,
> - 0x00044031, 0x00000000, 0xc0000414, 0x02a00000,
> - 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> - 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> - 0x80000901, 0x00010000, 0x00000000, 0x00000000,
> - }},
> - { .gen_ver = 0, .size = 48, .code = (const uint32_t []) {
> - 0x00020061, 0x01050000, 0x00000104, 0x00000000,
> - 0x00000069, 0x02058220, 0x02000024, 0x00000004,
> - 0x00000061, 0x02250220, 0x000000c4, 0x00000000,
> - 0x00030061, 0x04054220, 0x00000000, 0x00000000,
> - 0x00040461, 0x05050220, 0x00000104, 0x00000000,
> - 0x00010361, 0x04050220, 0x00220205, 0x00000000,
> - 0x00000061, 0x04454220, 0x00000000, 0x0000000f,
> - 0x00000061, 0x04850220, 0x000000a4, 0x00000000,
> - 0x00049031, 0x00000000, 0xc0000414, 0x02a00000,
> - 0x80000001, 0x00010000, 0x20000000, 0x00000000,
> - 0x80000001, 0x00010000, 0x30000000, 0x00000000,
> - 0x80000101, 0x00010000, 0x00000000, 0x00000000,
> - }}
> -};
> +#define MD5_SUM_IGA64_ASMS f5d2183439f5e7b18097c3c2c2957dbb
>
> struct iga64_template const iga64_code_read_a64_d32[] = {
> { .gen_ver = 2000, .size = 40, .code = (const uint32_t []) {
> diff --git a/lib/meson.build b/lib/meson.build
> index f5419805177f..c903aaa14bc9 100644
> --- a/lib/meson.build
> +++ b/lib/meson.build
> @@ -236,9 +236,30 @@ lib_version = vcs_tag(input : 'version.h.in', output : 'version.h',
> command : vcs_command )
>
> iga64_assembly_sources = [ 'gpgpu_shader.c', 'gpgpu_fill.c' ]
> +libiga64_asms = static_library('iga64_asms',
> + iga64_assembly_sources,
> + include_directories: inc,
> + dependencies : lib_deps,
> + c_args : [
> + '-DIGT_DATADIR="@0@"'.format(join_paths(prefix, datadir)),
> + '-DIGT_SRCDIR="@0@"'.format(srcdir),
> + '-DIGT_LOG_DOMAIN="gpu_codes"',
> + '-ffat-lto-objects' ]
> +)
> +
> +iga64_codes_targets = []
> +foreach f: iga64_assembly_sources
> + name = f.underscorify() + '.gen.iga64_codes.c'
> + iga64_codes_targets += custom_target(name,
> + input: [ f + '.gen.iga64_codes.c', libiga64_asms.extract_objects(f) ],
> + output: name,
> + depend_files: [ 'iga64_macros.h', generate_iga64_codes.path() ],
> + command: [ 'env', 'IGT_SRCDIR=@SOURCE_ROOT@', generate_iga64_codes, '-o', '@OUTPUT@', '-i', '@INPUT0@', '@INPUT1@' ],
> + console: true,
> + )
> +endforeach
>
> lib_intermediates = []
> -iga64_assembly_libs = []
> foreach f: lib_sources
> name = f.underscorify()
> lib = static_library('igt-' + name,
> @@ -249,25 +270,12 @@ foreach f: lib_sources
> '-DIGT_DATADIR="@0@"'.format(join_paths(prefix, datadir)),
> '-DIGT_SRCDIR="@0@"'.format(srcdir),
> '-DIGT_LOG_DOMAIN="@0@"'.format(f.split('.')[0]),
> - ] + (iga64_assembly_sources.contains(f) ? [ '-ffat-lto-objects' ] : []))
> + ])
>
> lib_intermediates += lib
> - if iga64_assembly_sources.contains(f)
> - iga64_assembly_libs += lib
> - endif
> endforeach
>
> -iga64_generated_codes = custom_target(
> - 'iga64_generated_codes.c',
> - input : [ 'iga64_generated_codes.c' ] + iga64_assembly_libs,
> - output : 'iga64_generated_codes.c',
> - depend_files : 'iga64_macros.h',
> - command : [ 'env', 'IGT_SRCDIR=@0@'.format(source_root), generate_iga64_codes, '-o', '@OUTPUT@', '-i', '@INPUT@' ]
> -)
> -
> -lib_intermediates += static_library('igt-iga64_generated_codes.c',
> - [ iga64_generated_codes, lib_version ]
> - )
> +lib_intermediates += static_library('iga64_gpu_codes', iga64_codes_targets)
>
> lib_igt_build = shared_library('igt',
> ['dummy.c'],
> diff --git a/scripts/generate_iga64_codes b/scripts/generate_iga64_codes
> index 1675ffe639b3..d196c7a016b7 100755
> --- a/scripts/generate_iga64_codes
> +++ b/scripts/generate_iga64_codes
> @@ -29,17 +29,17 @@ while getopts ':i:o:' opt; do
> case $opt in
> i) INPUT=$OPTARG;;
> o) OUTPUT=$OPTARG;;
> - ?) die "Usage: $0 -i pre-generated-iga64-file -o generated-iga64-file libs-with-iga64-assembly [...]"
> + ?) die "Usage: $0 -i pre-generated-iga64-file -o generated-iga64-file obj-with-iga64-assembly [...]"
> esac
> done
> -LIBS=${@:OPTIND}
> +OBJ=${@:OPTIND}
>
> # read all assemblies into ASMS array
> ASMS=()
> while read -d $'\0' asm; do
> test -z "$asm" && continue
> ASMS+=( "$asm" )
> -done < <(for f in $LIBS; do objcopy --dump-section .iga64_assembly=/dev/stdout $f.p/*.o /dev/null; done)
> +done < <(objcopy --dump-section .iga64_assembly=/dev/stdout $OBJ /dev/null)
>
> # check if we need to recompile - checksum difference and compiler present
> MD5_ASMS="$(md5sum <<< "$(< $IGT_SRCDIR/lib/iga64_macros.h) ${ASMS[@]}" | cut -b1-32)"
> @@ -47,13 +47,13 @@ MD5_PRE="$(grep -Po '(?<=^#define MD5_SUM_IGA64_ASMS )\S{32,32}' $INPUT 2>/dev/n
>
> if [ "$MD5_ASMS" = "$MD5_PRE" ]; then
> echo "iga64 assemblies not changed, reusing pre-compiled file $INPUT."
> - cp $INPUT $OUTPUT
> + cp -p $INPUT $OUTPUT
> exit 0
> fi
>
> type iga64 >/dev/null || {
> warn "WARNING: iga64 assemblies changed, but iga64 compiler not present, CHANGES will have no effect. Install iga64 (libigc-tools package) to re-compile code."
> - cp $INPUT $OUTPUT
> + cp -p $INPUT $OUTPUT
> exit 0
> }
>
> @@ -67,7 +67,7 @@ for gen in $GEN_VERSIONS; do
> gen_name="${gen#*:}"
> iga64 -p=$gen_name -d $WD/empty 2>/dev/null || {
> warn "WARNING: iga64 assemblies changed, but iga64 compiler does not support platform '$gen_name', CHANGES will have no effect. Update iga64 (libigc-tools package) to re-compile code."
> - cp $INPUT $OUTPUT
> + cp -p $INPUT $OUTPUT
> exit 0
> }
> done
> @@ -127,4 +127,4 @@ for asm in "${ASMS[@]}"; do
> echo -e "\t{ .gen_ver = $cur_ver, .size = $(dword_count "$cur_code"), .code = (const uint32_t []) {\n$cur_code\n\t}}\n};" >>$OUTPUT
> done
>
> -cp $OUTPUT $INPUT
> +cp -vp $OUTPUT $INPUT
>
> ---
> base-commit: f564652c85f6bf5e6e076526fe0de7c96dbbe0b2
> change-id: 20250528-refactor_gpgpu_asms-dd5fb1d78843
>
> Best regards,
> --
> Andrzej Hajda <andrzej.hajda at intel.com>
>
More information about the igt-dev
mailing list