Skip to contentSkip to conversation

ARMv8 32-bit compiler tunings

PATCH In review Cristian Bercaru 2 participants 4 messages Dec 4, 2015 – Dec 9, 2015

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARMv8 32-bit compiler tunings

Hello!

Starting with linux-yocto-4.1, Intel provides kernel support for its ARM v8 Axxia platforms, both on 32-bit and 64-bit.
Poky already contains arm 64-bit compiler tunings, so I added 32-bit tunings based on those for armv7.

The patch that I send you is a draft. Please review it and if it's ok I will return with thumb and big-endian tunings.

Thank you,
Cristian Bercaru (1):
add tunings for 32-bit ARMv8-a

 meta/conf/machine/include/arm/arch-armv7a.inc      |    1 -
 meta/conf/machine/include/arm/arch-armv8a.inc      |   65 ++++++++++++++++++++
 meta/conf/machine/include/arm/feature-arm-neon.inc |    3 -
 meta/conf/machine/include/arm/feature-arm-vfp.inc  |   59 ++++++++++++++----
 4 files changed, 112 insertions(+), 16 deletions(-)
 create mode 100644 meta/conf/machine/include/arm/arch-armv8a.inc
 delete mode 100644 meta/conf/machine/include/arm/feature-arm-neon.inc
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/1 add tunings for 32-bit ARMv8-a

This patch adds tunes for 32-bit armv8 platforms. The used can select the vector floating-point instruction set: vfpv3, vfpv4 or fp-armv8 and the Neon, crc and crypto extensions.

---
 meta/conf/machine/include/arm/arch-armv7a.inc      |    1 -
 meta/conf/machine/include/arm/arch-armv8a.inc      |   65 ++++++++++++++++++++
 meta/conf/machine/include/arm/feature-arm-neon.inc |    3 -
 meta/conf/machine/include/arm/feature-arm-vfp.inc  |   59 ++++++++++++++----
 4 files changed, 112 insertions(+), 16 deletions(-)
 create mode 100644 meta/conf/machine/include/arm/arch-armv8a.inc
 delete mode 100644 meta/conf/machine/include/arm/feature-arm-neon.inc
meta/conf/machine/include/arm/arch-armv7a.inc +0 -1
index d3b6f64..ac85fda 100644
@@ -6,7 +6,6 @@ TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "armv7a", " -march=armv7-a
 MACHINEOVERRIDES =. "${@bb.utils.contains("TUNE_FEATURES", "armv7a", "armv7a:", "" ,d)}"
 
 require conf/machine/include/arm/arch-armv6.inc
-require conf/machine/include/arm/feature-arm-neon.inc
 
 # Little Endian base configs
 AVAILTUNES += "armv7a armv7at armv7a-vfpv3d16 armv7at-vfpv3d16 armv7a-vfpv3 armv7at-vfpv3 armv7a-neon armv7at-neon"
meta/conf/machine/include/arm/arch-armv8a.inc new file +65 -0

65 lines not on this page · raw message

meta/conf/machine/include/arm/feature-arm-neon.inc deleted +0 -3
deleted file mode 100644
index e8b2b85..0000000
@@ -1,3 +0,0 @@
-TUNEVALID[neon] = "Enable Neon SIMD accelerator unit."
-TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "neon", bb.utils.contains("TUNE_FEATURES", "vfpv4", " -mfpu=neon-vfpv4", " -mfpu=neon", d), "" , d)}"
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "neon", "-neon", "", d)}"
meta/conf/machine/include/arm/feature-arm-vfp.inc +47 -12

64 lines not on this page · raw message

Signature · 1 line
1.7.9.5
#2 lore raw

Re: [PATCH] add tunings for 32-bit ARMv8-a

Quoted 3 lines · Cristian Bercaru

This patch adds tunes for 32-bit armv8 platforms. The used can select the vector floating-point instruction set: vfpv3, vfpv4 or fp-armv8 and the Neon, crc and crypto extensions.

Wrong ML, it belongs to oe-core ML.

Please check this patchset:
http://git.openembedded.org/openembedded-core-contrib/log/?h=jansa/tune3
and test script provided in:
http://git.openembedded.org/openembedded-core-contrib/log/?h=jansa/tune2-test

It was sent couple months ago, you can find the discussion in various threads, e.g.:
http://lists.openembedded.org/pipermail/openembedded-core/2015-October/111318.html
http://lists.openembedded.org/pipermail/openembedded-core/2015-November/112907.html

It fixes some of these issues and simplifies implementing the rest.

Quoted 114 lines
---
 meta/conf/machine/include/arm/arch-armv7a.inc      |    1 -
 meta/conf/machine/include/arm/arch-armv8a.inc      |   65 ++++++++++++++++++++
 meta/conf/machine/include/arm/feature-arm-neon.inc |    3 -
 meta/conf/machine/include/arm/feature-arm-vfp.inc  |   59 ++++++++++++++----
 4 files changed, 112 insertions(+), 16 deletions(-)
 create mode 100644 meta/conf/machine/include/arm/arch-armv8a.inc
 delete mode 100644 meta/conf/machine/include/arm/feature-arm-neon.inc
meta/conf/machine/include/arm/arch-armv7a.inc +0 -1
index d3b6f64..ac85fda 100644
@@ -6,7 +6,6 @@ TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "armv7a", " -march=armv7-a
 MACHINEOVERRIDES =. "${@bb.utils.contains("TUNE_FEATURES", "armv7a", "armv7a:", "" ,d)}"
 
 require conf/machine/include/arm/arch-armv6.inc
-require conf/machine/include/arm/feature-arm-neon.inc
 
 # Little Endian base configs
 AVAILTUNES += "armv7a armv7at armv7a-vfpv3d16 armv7at-vfpv3d16 armv7a-vfpv3 armv7at-vfpv3 armv7a-neon armv7at-neon"
meta/conf/machine/include/arm/arch-armv8a.inc new file +65 -0
new file mode 100644
index 0000000..1b062b8
@@ -0,0 +1,65 @@
+DEFAULTTUNE ?= "armv8a"
+
+TUNEVALID[armv8a] = "Enable instructions for ARMv8-a"
+TUNEVALID[crc] = "Enable CRC instrucitons for ARMv8-a"
+TUNECONFLICTS[armv8a] = "armv4 armv5 armv6 armv7 armv7a"
+TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "armv8a", " -march=armv8-a", "", d)}"
+TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "crc", "+crc", "", d)}"
+MACHINEOVERRIDES =. "${@bb.utils.contains("TUNE_FEATURES", "armv8a", "armv8a:", "" ,d)}"
+
+require conf/machine/include/arm/arch-armv7a.inc
+
+# Little Endian base configs
+AVAILTUNES += "armv8a armv8a-vfpv3 armv8a-neon armv8a-vfpv4 armv8a-neon-vfpv4 armv8a-fp-armv8 armv8a-neon-fp-armv8 armv8a-crypto-neon-fp-armv8"
+ARMPKGARCH_tune-armv8a ?= "armv8a"
+ARMPKGARCH_tune-armv8a-vfpv3 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-neon ?= "armv8a"
+ARMPKGARCH_tune-armv8a-vfpv4 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-neon-vfpv4 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-fp-armv8 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-neon-fp-armv8 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crypto-neon-fp-armv8 ?= "armv8a"
+TUNE_FEATURES_tune-armv8a ?= "arm armv8a vfp"
+TUNE_FEATURES_tune-armv8a-vfpv3 ?= "${TUNE_FEATURES_tune-armv8a} vfpv3"
+TUNE_FEATURES_tune-armv8a-neon ?= "${TUNE_FEATURES_tune-armv8a} neon"
+TUNE_FEATURES_tune-armv8a-vfpv4 ?= "${TUNE_FEATURES_tune-armv8a} vfpv4"
+TUNE_FEATURES_tune-armv8a-neon-vfpv4 ?= "${TUNE_FEATURES_tune-armv8a} neon vfpv4"
+TUNE_FEATURES_tune-armv8a-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a} fp-armv8"
+TUNE_FEATURES_tune-armv8a-neon-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a} neon fp-armv8"
+TUNE_FEATURES_tune-armv8a-crypto-neon-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a} crypto neon fp-armv8"
+PACKAGE_EXTRA_ARCHS_tune-armv8a = "${PACKAGE_EXTRA_ARCHS_tune-armv7a} armv8a armv8a-vfp"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-vfpv3 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a} armv8a-vfp-vfpv3"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-neon = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-vfpv3} armv8a-vfp-neon"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-vfpv4 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-vfpv3} armv8a-vfp-vfpv4"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-neon-vfpv4 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-vfpv4} armv8a-vfp-neon armv8a-vfp-neon-vfpv4"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a} armv8a-vfp-fp-armv8"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-neon-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-fp-armv8} armv8a-vfp-neon-fp-armv8"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crypto-neon-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-neon-fp-armv8} armv8a-vfp-crypto-neon-fp-armv8"
+
+# CRC configs
+AVAILTUNES += "armv8a-crc armv8a-crc-vfpv3 armv8a-crc-neon armv8a-crc-vfpv4 armv8a-crc-neon-vfpv4 armv8a-crc-fp-armv8 armv8a-crc-neon-fp-armv8 armv8a-crc-crypto-neon-fp-armv8"
+ARMPKGARCH_tune-armv8a-crc ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-vfpv3 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-neon ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-vfpv4 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-neon-vfpv4 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-fp-armv8 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-neon-fp-armv8 ?= "armv8a"
+ARMPKGARCH_tune-armv8a-crc-crypto-neon-fp-armv8 ?= "armv8a"
+TUNE_FEATURES_tune-armv8a-crc ?= "${TUNE_FEATURES_tune-armv8a} crc"
+TUNE_FEATURES_tune-armv8a-crc-vfpv3 ?= "${TUNE_FEATURES_tune-armv8a-vfpv3} crc"
+TUNE_FEATURES_tune-armv8a-crc-neon ?= "${TUNE_FEATURES_tune-armv8a-neon} crc"
+TUNE_FEATURES_tune-armv8a-crc-vfpv4 ?= "${TUNE_FEATURES_tune-armv8a-vfpv4} crc"
+TUNE_FEATURES_tune-armv8a-crc-neon-vfpv4 ?= "${TUNE_FEATURES_tune-armv8a-neon-vfpv4} crc"
+TUNE_FEATURES_tune-armv8a-crc-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a-fp-armv8} crc"
+TUNE_FEATURES_tune-armv8a-crc-neon-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a-neon-fp-armv8} crc"
+TUNE_FEATURES_tune-armv8a-crc-crypto-neon-fp-armv8 ?= "${TUNE_FEATURES_tune-armv8a-crypto-neon-fp-armv8} crc"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc = "${PACKAGE_EXTRA_ARCHS_tune-armv8a} armv8a-crc-vfp"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-vfpv3 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc} armv8a-vfp-vfpv3 armv8a-crc-vfp-vfpv3"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-neon = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-vfpv3} armv8a-vfp-neon armv8a-crc-vfp-neon"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-vfpv4 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-vfpv3} armv8a-vfp-vfpv4 armv8a-crc-vfp-vfpv4"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-neon-vfpv4 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-vfpv4} armv8a-vfp-neon armv8a-crc-vfp-neon armv8a-vfp-neon-vfpv4 armv8a-crc-vfp-neon-vfpv4"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc} armv8a-vfp-fp-armv8 armv8a-vfp-fp-crc-armv8"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-neon-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-fp-armv8} armv8a-vfp-neon-fp-armv8 armv8a-crc-vfp-neon-fp-armv8"
+PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-crypto-neon-fp-armv8 = "${PACKAGE_EXTRA_ARCHS_tune-armv8a-crc-neon-fp-armv8} armv8a-vfp-crypto-neon-fp-armv8 armv8a-crc-vfp-crypto-neon-fp-armv8"
+
meta/conf/machine/include/arm/feature-arm-neon.inc deleted +0 -3
deleted file mode 100644
index e8b2b85..0000000
@@ -1,3 +0,0 @@
-TUNEVALID[neon] = "Enable Neon SIMD accelerator unit."
-TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "neon", bb.utils.contains("TUNE_FEATURES", "vfpv4", " -mfpu=neon-vfpv4", " -mfpu=neon", d), "" , d)}"
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "neon", "-neon", "", d)}"
meta/conf/machine/include/arm/feature-arm-vfp.inc +0 -5
index 3dfbeac..12a4e81 100644
@@ -1,17 +1,52 @@
 TUNEVALID[vfp] = "Enable Vector Floating Point (vfp) unit."
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "vfp", "-vfp", "" ,d)}"
-
 TUNEVALID[vfpv3d16] = "Enable Vector Floating Point Version 3 with 16 registers (vfpv3-d16) unit."
-TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "vfpv3d16", " -mfpu=vfpv3-d16", "", d)}"
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "vfpv3d16", "-vfpv3d16", "" ,d)}"
-

meta/conf/machine/include/arm/feature-arm-vfp.inc line 7

Quoted 10 lines
meta/conf/machine/include/arm/feature-arm-vfp.inc +3 -5
 TUNEVALID[vfpv3] = "Enable Vector Floating Point Version 3 with 32 registers (vfpv3) unit."
-TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "vfpv3", " -mfpu=vfpv3", "", d)}"
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "vfpv3", "-vfpv3", "" ,d)}"
-
 TUNEVALID[vfpv4] = "Enable Vector Floating Point Version 4 (vfpv4) unit."
-ARMPKGSFX_FPU .= "${@bb.utils.contains("TUNE_FEATURES", "vfpv4", "-vfpv4", "" ,d)}"
-
+TUNEVALID[fp-armv8] = "Enable ARMv8 Vector Floating Point unit."
+TUNEVALID[neon] = "Enable vfpv3 and Neon SIMD accelerator unit."
+TUNEVALID[crypto] = "Enable ARMv8 crypto extension."

feature-arm-vfp.inc is included in armv5.inc, so it makes a lot of these features added to TUNEVALID actually invalid for armv[56] MACHINEs.

Quoted 55 lines
meta/conf/machine/include/arm/feature-arm-vfp.inc +44 -2
 TUNEVALID[callconvention-hard] = "Enable EABI hard float call convention, requires VFP."
-TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "vfp", bb.utils.contains("TUNE_FEATURES", "callconvention-hard", " -mfloat-abi=hard", " -mfloat-abi=softfp", d), "" ,d)}"
-ARMPKGSFX_EABI .= "${@bb.utils.contains("TUNE_FEATURES", [ "callconvention-hard", "vfp" ], "hf", "", d)}"
+
+python () {
+	if bb.utils.contains("TUNE_FEATURES", "crc", True, False, d):
+		d.appendVar("ARMPKGSFX_FPU", "-crc")
+
+	if bb.utils.contains("TUNE_FEATURES", "vfp", True, False, d):
+		d.appendVar("ARMPKGSFX_FPU", "-vfp")
+
+	if bb.utils.contains("TUNE_FEATURES", "vfpv3d16", True, False, d):
+		d.appendVar("TUNE_CCARGS", " -mfpu=vfpv3-d16")
+		d.appendVar("ARMPKGSFX_FPU", "-vfpv3d16")
+
+	if bb.utils.contains("TUNE_FEATURES", "vfpv3", True, False, d):
+			d.appendVar("TUNE_CCARGS", " -mfpu=vfpv3")
+			d.appendVar("ARMPKGSFX_FPU", "-vfpv3")
+	elif bb.utils.contains("TUNE_FEATURES", "vfpv4", True, False, d):
+		if bb.utils.contains("TUNE_FEATURES", "neon", True, False, d):
+			d.appendVar("TUNE_CCARGS", " -mfpu=neon-vfpv4")
+			d.appendVar("ARMPKGSFX_FPU", "-neon-vfpv4")
+		else:
+			d.appendVar("TUNE_CCARGS", " -mfpu=vfpv4")
+			d.appendVar("ARMPKGSFX_FPU", "-vfpv4")
+	elif bb.utils.contains("TUNE_FEATURES", "fp-armv8", True, False, d):
+		if bb.utils.contains("TUNE_FEATURES", "neon", True, False, d):
+			if bb.utils.contains("TUNE_FEATURES", "crypto", True, False, d):
+				d.appendVar("TUNE_CCARGS", " -mfpu=crypto-neon-fp-armv8")
+				d.appendVar("ARMPKGSFX_FPU", "-crypto-neon-fp-armv8")
+			else:
+				d.appendVar("TUNE_CCARGS", " -mfpu=neon-fp-armv8")
+				d.appendVar("ARMPKGSFX_FPU", "-neon-fp-armv8")
+		else:
+			d.appendVar("TUNE_CCARGS", " -mfpu=fp-armv8")
+			d.appendVar("ARMPKGSFX_FPU", "-fp-armv8")
+	elif bb.utils.contains("TUNE_FEATURES", "neon", True, False, d):
+		d.appendVar("TUNE_CCARGS", " -mfpu=neon")
+		d.appendVar("ARMPKGSFX_FPU", "-neon")
+
+	if bb.utils.contains("TUNE_FEATURES", "callconvention-hard", True, False, d):
+		d.appendVar("TUNE_CCARGS", " -mfloat-abi=hard")
+	else:
+		d.appendVar("TUNE_CCARGS", " -mfloat-abi=softfp")
+	if bb.utils.contains("TUNE_FEATURES", [ "vfp", "callconvention-hard" ], True, False, d):
+		d.appendVar("ARMPKGSFX_EABI", "hf")
+}
Signature · 7 lines
1.7.9.5

-- 
_______________________________________________
poky mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/poky
Signature · 1 line
Martin 'JaMa' Jansa     jabber: [email protected]
#3 lore raw
In reply to Cristian Bercaru

So, what do you guys think about my patch? Is the armv8a support a valuable addition to poky?

Cristian

Quoted 22 lines · Cristian Bercaru

Hello!

Starting with linux-yocto-4.1, Intel provides kernel support for its ARM v8 Axxia platforms, both on 32-bit and 64-bit.
Poky already contains arm 64-bit compiler tunings, so I added 32-bit tunings based on those for armv7.

The patch that I send you is a draft. Please review it and if it's ok I will return with thumb and big-endian tunings.

Thank you,
Cristian Bercaru (1):
add tunings for 32-bit ARMv8-a

  meta/conf/machine/include/arm/arch-armv7a.inc      |    1 -
  meta/conf/machine/include/arm/arch-armv8a.inc      |   65 ++++++++++++++++++++
  meta/conf/machine/include/arm/feature-arm-neon.inc |    3 -
  meta/conf/machine/include/arm/feature-arm-vfp.inc  |   59 ++++++++++++++----
  4 files changed, 112 insertions(+), 16 deletions(-)

create mode 100644 meta/conf/machine/include/arm/arch-armv8a.inc delete mode 100644 meta/conf/machine/include/arm/feature-arm-neon.inc