mirror of
https://github.com/amd/blis.git
synced 2026-07-13 10:37:34 +00:00
05e837d176
Implement zen6 cpuid and arch changes, and add zen6 as a separate BLIS sub-configuration and code path within amdzen configuration family. Currently all optimization choices are copies of zen5 sub-configuration. AMD-Internal: [CPUPL-7162]
180 lines
8.8 KiB
Makefile
180 lines
8.8 KiB
Makefile
#
|
|
#
|
|
# BLIS
|
|
# An object-based framework for developing high-performance BLAS-like
|
|
# libraries.
|
|
#
|
|
# Copyright (C) 2026, Advanced Micro Devices, Inc. All rights reserved.
|
|
#
|
|
# Redistribution and use in source and binary forms, with or without
|
|
# modification, are permitted provided that the following conditions are
|
|
# met:
|
|
# - Redistributions of source code must retain the above copyright
|
|
# notice, this list of conditions and the following disclaimer.
|
|
# - Redistributions in binary form must reproduce the above copyright
|
|
# notice, this list of conditions and the following disclaimer in the
|
|
# documentation and/or other materials provided with the distribution.
|
|
# - Neither the name(s) of the copyright holder(s) nor the names of its
|
|
# contributors may be used to endorse or promote products derived
|
|
# from this software without specific prior written permission.
|
|
#
|
|
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
# HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
#
|
|
#
|
|
|
|
# FLAGS that are specific to the 'zen6' architecture are added here.
|
|
# FLAGS that are common for all the AMD architectures are present in
|
|
# config/zen/amd_config.mk.
|
|
|
|
# Declare the name of the current configuration and add it to the
|
|
# running list of configurations included by common.mk.
|
|
THIS_CONFIG := zen6
|
|
#CONFIGS_INCL += $(THIS_CONFIG)
|
|
|
|
# Include file containing common flags for all AMD architectures
|
|
AMD_CONFIG_FILE := amd_config.mk
|
|
AMD_CONFIG_PATH := $(BASE_SHARE_PATH)/config/zen
|
|
-include $(AMD_CONFIG_PATH)/$(AMD_CONFIG_FILE)
|
|
|
|
ifeq ($(CC_VENDOR),gcc)
|
|
ifeq ($(shell test $(CC_MAJOR) -ge 16; echo $$?),0)
|
|
# gcc 16.0 or later
|
|
CKVECFLAGS += -march=znver6
|
|
CRVECFLAGS += -march=znver6
|
|
# Update CKLPOPTFLAGS for gcc to use O3 optimization without
|
|
# -ftree-pre and -ftree-partial-pre flag. These flag results
|
|
# in suboptimal code generation for instrinsic based kernels.
|
|
# The -ftree-loop-vectorize results in inefficient code gen
|
|
# for amd optimized l1 kernels based on instrinsics.
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 14; echo $$?),0)
|
|
# gcc 14.0 or later
|
|
CKVECFLAGS += -march=znver5 -mavx512fp16
|
|
CRVECFLAGS += -march=znver5
|
|
# Update CKLPOPTFLAGS for gcc to use O3 optimization without
|
|
# -ftree-pre and -ftree-partial-pre flag. These flag results
|
|
# in suboptimal code generation for instrinsic based kernels.
|
|
# The -ftree-loop-vectorize results in inefficient code gen
|
|
# for amd optimized l1 kernels based on instrinsics.
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 13; echo $$?),0)
|
|
# gcc 13.0 or later
|
|
CKVECFLAGS += -march=znver4 -mavx512fp16
|
|
CRVECFLAGS += -march=znver4
|
|
# Update CKLPOPTFLAGS for gcc to use O3 optimization without
|
|
# -ftree-pre and -ftree-partial-pre flag. These flag results
|
|
# in suboptimal code generation for instrinsic based kernels.
|
|
# The -ftree-loop-vectorize results in inefficient code gen
|
|
# for amd optimized l1 kernels based on instrinsics.
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 12; echo $$?),0)
|
|
# gcc 12.0 or later
|
|
CKVECFLAGS += -march=znver3 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512bf16 -mavx512vbmi -mavx512fp16
|
|
CRVECFLAGS += -march=znver3
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 11; echo $$?),0)
|
|
# gcc 11.0 or later
|
|
CKVECFLAGS += -march=znver3 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512bf16 -mavx512vbmi
|
|
CRVECFLAGS += -march=znver3
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 9; echo $$?),0)
|
|
# gcc 9.0 or later
|
|
CKVECFLAGS += -march=znver2 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512vbmi
|
|
CRVECFLAGS += -march=znver2
|
|
CKLPOPTFLAGS += -fno-tree-partial-pre -fno-tree-pre -fno-tree-loop-vectorize
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 8; echo $$?),0)
|
|
# gcc 8.0 or later
|
|
CKVECFLAGS += -march=znver1 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512vbmi
|
|
CRVECFLAGS += -march=znver1
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 7; echo $$?),0)
|
|
# gcc 7.0 or later
|
|
CKVECFLAGS += -march=znver1 -mavx512f -mavx512dq -mavx512bw -mavx512vl
|
|
CRVECFLAGS += -march=znver1
|
|
else
|
|
# If gcc is older than 7.0.0 but at least 6.1.0, then we can use -march=znver1
|
|
# as the fallback option.
|
|
CKVECFLAGS += -march=znver1 -mno-avx256-split-unaligned-store
|
|
CRVECFLAGS += -march=znver1 -mno-avx256-split-unaligned-store
|
|
endif
|
|
endif # gcc
|
|
|
|
ifeq ($(CC_VENDOR),clang)
|
|
# AOCC clang has various formats for the version line
|
|
|
|
# AOCC.LLVM.2.0.0.B191.2019_07_19 clang version 8.0.0 (CLANG: Jenkins AOCC_2_0_0-Build#191) (based on LLVM AOCC.LLVM.2.0.0.B191.2019_07_19)
|
|
# AOCC.LLVM.2.1.0.B1030.2019_11_12 clang version 9.0.0 (CLANG: Build#1030) (based on LLVM AOCC.LLVM.2.1.0.B1030.2019_11_12)
|
|
# AMD clang version 10.0.0 (CLANG: AOCC_2.2.0-Build#93 2020_06_25) (based on LLVM Mirror.Version.10.0.0)
|
|
# AMD clang version 11.0.0 (CLANG: AOCC_2.3.0-Build#85 2020_11_10) (based on LLVM Mirror.Version.11.0.0)
|
|
# AMD clang version 12.0.0 (CLANG: AOCC_3.0.0-Build#2 2020_11_05) (based on LLVM Mirror.Version.12.0.0)
|
|
# AMD clang version 14.0.0 (CLANG: AOCC_4.0.0-Build#98 2022_06_15) (based on LLVM Mirror.Version.14.0.0)
|
|
ifneq ($(IS_AOCC),)
|
|
ifeq ($(shell test $(AOCC_MAJOR) -ge 6; echo $$?),0)
|
|
# AOCC version 6.x or later (6.0.0+)
|
|
CKVECFLAGS += -march=znver6 -falign-loops=64
|
|
CRVECFLAGS += -march=znver6
|
|
else ifeq ($(shell test $(AOCC_MAJOR) -ge 5; echo $$?),0)
|
|
# AOCC version 5.x or later (5.0.0+)
|
|
CKVECFLAGS += -march=znver5 -mavx512fp16 -falign-loops=64
|
|
CRVECFLAGS += -march=znver5
|
|
else ifeq ($(shell test $(AOCC_MAJOR) -ge 4; echo $$?),0)
|
|
# AOCC version 4.x (4.0.0+)
|
|
CKVECFLAGS += -march=znver4 -mavx512fp16 -falign-loops=64
|
|
CRVECFLAGS += -march=znver4
|
|
else ifeq ($(shell test $(AOCC_MAJOR) -ge 3; echo $$?),0)
|
|
# AOCC version 3.x (3.0.0+)
|
|
CKVECFLAGS += -march=znver3 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512bf16 -mavx512vbmi -falign-loops=64
|
|
CRVECFLAGS += -march=znver3
|
|
else ifeq ($(shell test $(AOCC_MAJOR) -ge 2; echo $$?),0)
|
|
# AOCC version 2.x (2.0.0+)
|
|
CKVECFLAGS += -march=znver2 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512vbmi
|
|
CRVECFLAGS += -march=znver2
|
|
endif
|
|
else
|
|
# No AOCC version detected, use LLVM clang version
|
|
ifeq ($(shell test $(CC_MAJOR) -ge 23; echo $$?),0)
|
|
# LLVM clang 23.0 or later
|
|
CKVECFLAGS += -march=znver6 -falign-loops=64
|
|
CRVECFLAGS += -march=znver6
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 19; echo $$?),0)
|
|
# LLVM clang 19.0 or later
|
|
CKVECFLAGS += -march=znver5 -falign-loops=64
|
|
CRVECFLAGS += -march=znver5
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 16; echo $$?),0)
|
|
# LLVM clang 16.0 or later
|
|
CKVECFLAGS += -march=znver4 -falign-loops=64
|
|
CRVECFLAGS += -march=znver4
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 13; echo $$?),0)
|
|
# LLVM clang 13.0 or later
|
|
CKVECFLAGS += -march=znver3 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512bf16 -mavx512vbmi -falign-loops=64
|
|
CRVECFLAGS += -march=znver3
|
|
else ifeq ($(shell test $(CC_MAJOR) -ge 9; echo $$?),0)
|
|
# LLVM clang 9.0 or later
|
|
CKVECFLAGS += -march=znver2 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512bf16 -mavx512vbmi -falign-loops=64
|
|
CRVECFLAGS += -march=znver2
|
|
else
|
|
# LLVM clang < 9.0
|
|
CKVECFLAGS += -march=znver1 -mavx512f -mavx512dq -mavx512bw -mavx512vl -mavx512vnni -mavx512vbmi -falign-loops=64
|
|
CRVECFLAGS += -march=znver1
|
|
endif
|
|
endif
|
|
endif # clang
|
|
|
|
# Flags specific to reference kernels.
|
|
CROPTFLAGS := $(CKOPTFLAGS)
|
|
CRVECFLAGS := $(CKVECFLAGS)
|
|
|
|
# Store all of the variables here to new variables containing the
|
|
# configuration name.
|
|
$(eval $(call store-make-defs,$(THIS_CONFIG)))
|
|
|