Extend x86 to have different arch variants

Author: Negreanu Marius Adrian <adrian.m.negreanu@intel.com>
Author: Andrew Boie <andrew.p.boie@intel.com>
Author: Daniel Leung <daniel.leung@intel.com>

Currently, x86 target only has generic i686 and x86-atom
as arch variants. This patch adds the ability to have
more than two arch variants. Defining a new arch variant
is similiar to ARM targets, by adding a new file in
core/combo/arch/x86. These files also define what
capabilities the targeting CPU has (e.g. having SSE2,
SSE3, etc.).

We define arch variants for Sandy Bridge, Ivy Bridge,
Haswell; upcoming arches can be easily added to this
set with future patches.

Change-Id: Iafbce10d205e860738db4a216ff603f9a84d7311
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>

6 files changed, 96 insertions, 19 deletions

diff --git a/core/combo/TARGET_linux-x86.mk b/core/combo/TARGET_linux-x86.mk
index 159fbe1..5859bd4 100644
--- a/core/combo/TARGET_linux-x86.mk
+++ b/core/combo/TARGET_linux-x86.mk
@@ -131,7 +131,8 @@ TARGET_GLOBAL_CFLAGS += \
 			-fstrict-aliasing \
 			-funswitch-loops \
 			-funwind-tables \
-			-fstack-protector
+			-fstack-protector \
+			-m32
 
 android_config_h := $(call select-android-config-h,target_linux-x86)
 TARGET_ANDROID_CONFIG_CFLAGS := -include $(android_config_h) -I $(dir $(android_config_h))
@@ -141,23 +142,37 @@ TARGET_GLOBAL_CFLAGS += $(TARGET_ANDROID_CONFIG_CFLAGS)
 TARGET_GLOBAL_CPPFLAGS += \
 			-fno-use-cxa-atexit
 
-# XXX: Our toolchain is normally configured to always set these flags by default
-# however, there have been reports that this is sometimes not the case. So make
-# them explicit here unless we have the time to carefully check it
-#
-TARGET_GLOBAL_CFLAGS += -mstackrealign -msse3 -mfpmath=sse -m32
+TARGET_GLOBAL_CFLAGS += $(arch_variant_cflags)
 
-# XXX: These flags should not be defined here anymore. Instead, the Android.mk
-# of the modules that depend on these features should instead check the
-# corresponding macros (e.g. ARCH_X86_HAVE_SSE2 and ARCH_X86_HAVE_SSSE3)
-# Keep them here until this is all cleared up.
-#
+ifeq ($(ARCH_X86_HAVE_MMX),true)
+    TARGET_GLOBAL_CFLAGS += -DUSE_MMX -mmmx
+endif
+ifeq ($(ARCH_X86_HAVE_SSE),true)
+    TARGET_GLOBAL_CFLAGS += -DUSE_SSE -msse
+endif
 ifeq ($(ARCH_X86_HAVE_SSE2),true)
-TARGET_GLOBAL_CFLAGS += -DUSE_SSE2
+    TARGET_GLOBAL_CFLAGS += -DUSE_SSE2 -msse2
+endif
+ifeq ($(ARCH_X86_HAVE_SSE3),true)
+    TARGET_GLOBAL_CFLAGS += -DUSE_SSE3 -msse3
 endif
-
 ifeq ($(ARCH_X86_HAVE_SSSE3),true)   # yes, really SSSE3, not SSE3!
-TARGET_GLOBAL_CFLAGS += -DUSE_SSSE3
+    TARGET_GLOBAL_CFLAGS += -DUSE_SSSE3 -mssse3
+endif
+ifeq ($(ARCH_X86_HAVE_SSE4),true)
+    TARGET_GLOBAL_CFLAGS += -msse4
+endif
+ifeq ($(ARCH_X86_HAVE_SSE4_1),true)
+    TARGET_GLOBAL_CFLAGS += -msse4.1
+endif
+ifeq ($(ARCH_X86_HAVE_SSE4_2),true)
+    TARGET_GLOBAL_CFLAGS += -msse4.2
+endif
+ifeq ($(ARCH_X86_HAVE_AVX),true)
+    TARGET_GLOBAL_CFLAGS += -mavx
+endif
+ifeq ($(ARCH_X86_HAVE_AES_NI),true)
+    TARGET_GLOBAL_CFLAGS += -maes
 endif
 
 # XXX: This flag is probably redundant. I believe our toolchain always sets
diff --git a/core/combo/arch/x86/haswell.mk b/core/combo/arch/x86/haswell.mk
new file mode 100644
index 0000000..5cbb3b8
--- /dev/null
+++ b/core/combo/arch/x86/haswell.mk
@@ -0,0 +1,20 @@
+# Configuration for Linux on x86.
+# Generating binaries for Haswell processors.
+#
+ARCH_X86_HAVE_MMX    := true
+ARCH_X86_HAVE_SSE    := true
+ARCH_X86_HAVE_SSE2   := true
+ARCH_X86_HAVE_SSE3   := true
+ARCH_X86_HAVE_SSSE3  := true
+ARCH_X86_HAVE_SSE4   := true
+ARCH_X86_HAVE_SSE4_1 := true
+ARCH_X86_HAVE_SSE4_2 := true
+ARCH_X86_HAVE_AES_NI := true
+ARCH_X86_HAVE_AVX    := true
+
+# CFLAGS for this arch
+arch_variant_cflags := \
+	-march=core-avx2 \
+	-mstackrealign \
+	-mfpmath=sse \
+
diff --git a/core/combo/arch/x86/ivybridge.mk b/core/combo/arch/x86/ivybridge.mk
new file mode 100644
index 0000000..c0f8d89
--- /dev/null
+++ b/core/combo/arch/x86/ivybridge.mk
@@ -0,0 +1,20 @@
+# Configuration for Linux on x86.
+# Generating binaries for Ivy Bridge processors.
+#
+ARCH_X86_HAVE_MMX    := true
+ARCH_X86_HAVE_SSE    := true
+ARCH_X86_HAVE_SSE2   := true
+ARCH_X86_HAVE_SSE3   := true
+ARCH_X86_HAVE_SSSE3  := true
+ARCH_X86_HAVE_SSE4   := true
+ARCH_X86_HAVE_SSE4_1 := true
+ARCH_X86_HAVE_SSE4_2 := true
+ARCH_X86_HAVE_AES_NI := true
+ARCH_X86_HAVE_AVX    := true
+
+# CFLAGS for this arch
+arch_variant_cflags := \
+	-march=core-avx-i \
+	-mstackrealign \
+	-mfpmath=sse \
+
diff --git a/core/combo/arch/x86/sandybridge.mk b/core/combo/arch/x86/sandybridge.mk
new file mode 100644
index 0000000..8d5e609
--- /dev/null
+++ b/core/combo/arch/x86/sandybridge.mk
@@ -0,0 +1,18 @@
+# Configuration for Linux on x86.
+# Generating binaries for SandyBridge processors.
+#
+ARCH_X86_HAVE_MMX    := true
+ARCH_X86_HAVE_SSE    := true
+ARCH_X86_HAVE_SSE2   := true
+ARCH_X86_HAVE_SSE3   := true
+ARCH_X86_HAVE_SSSE3  := true
+ARCH_X86_HAVE_SSE4_1 := true
+ARCH_X86_HAVE_SSE4_2 := true
+ARCH_X86_HAVE_AVX    := true
+
+# CFLAGS for this arch
+arch_variant_cflags := \
+	-march=corei7-avx \
+	-mstackrealign \
+	-mfpmath=sse \
+
diff --git a/core/combo/arch/x86/x86-atom.mk b/core/combo/arch/x86/x86-atom.mk
index 85998e7..1ea4eea 100644
--- a/core/combo/arch/x86/x86-atom.mk
+++ b/core/combo/arch/x86/x86-atom.mk
@@ -8,11 +8,13 @@ ARCH_X86_HAVE_MMX   := true
 ARCH_X86_HAVE_SSE   := true
 ARCH_X86_HAVE_SSE2  := true
 ARCH_X86_HAVE_SSE3  := true
-
 ARCH_X86_HAVE_SSSE3 := true
 ARCH_X86_HAVE_MOVBE := true
 ARCH_X86_HAVE_POPCNT := false   # popcnt is not supported by current Atom CPUs
 
-# This flag is used to enabled Atom-specific optimizations with our toolchain
-#
-TARGET_GLOBAL_CFLAGS += -march=atom
+# CFLAGS for this arch
+arch_variant_cflags := \
+	-march=atom \
+	-mstackrealign \
+	-mfpmath=sse \
+
diff --git a/core/combo/arch/x86/x86.mk b/core/combo/arch/x86/x86.mk
index 476da45..a62d86c 100644
--- a/core/combo/arch/x86/x86.mk
+++ b/core/combo/arch/x86/x86.mk
@@ -32,4 +32,6 @@ ARCH_X86_HAVE_POPCNT := false
 # not always work as intended, so keep it unless we have the time to check
 # everything properly.
 
-TARGET_GLOBAL_CFLAGS += -march=i686
+arch_variant_cflags := \
+    -march=i686 \
+