From 4648c3661fa68c864eb53c0594b1e9f38d1de224 Mon Sep 17 00:00:00 2001
From: Devin Matthews <damatthews@smu.edu>
Date: Wed, 26 Oct 2022 18:10:26 -0500
Subject: [PATCH 1/2] Fix auto-detection of firestorm (Apple M1).

---
 frame/base/bli_cpuid.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/frame/base/bli_cpuid.c b/frame/base/bli_cpuid.c
index 527db1f5d7..d967cc05d6 100644
--- a/frame/base/bli_cpuid.c
+++ b/frame/base/bli_cpuid.c
@@ -781,7 +781,7 @@ uint32_t bli_cpuid_query
 		if ( bli_cpuid_has_features( ecx, FEATURE_MASK_AVX   ) ) *features |= FEATURE_AVX;
 		if ( bli_cpuid_has_features( ecx, FEATURE_MASK_FMA3  ) ) *features |= FEATURE_FMA3;
 
-		// Check whether the hardware supports xsave/xrestor/xsetbv/xgetbv AND 
+		// Check whether the hardware supports xsave/xrestor/xsetbv/xgetbv AND
 		// support for these is enabled by the OS. If so, then we proceed with
 		// checking that various register-state saving features are available.
 		if ( bli_cpuid_has_features( ecx, FEATURE_MASK_XGETBV ) )
@@ -813,7 +813,7 @@ uint32_t bli_cpuid_query
 
 			// The OS can manage the state of 512-bit zmm (AVX-512) registers
 			// only if the xcr[7:5] bits are set. If they are not set, then
-			// clear all feature bits related to AVX-512. 
+			// clear all feature bits related to AVX-512.
 			if ( !bli_cpuid_has_features( eax, XGETBV_MASK_XMM |
 				                               XGETBV_MASK_YMM |
 				                               XGETBV_MASK_ZMM ) )
@@ -829,7 +829,7 @@ uint32_t bli_cpuid_query
 
 			// The OS can manage the state of 256-bit ymm (AVX) registers
 			// only if the xcr[2] bit is set. If it is not set, then
-			// clear all feature bits related to AVX. 
+			// clear all feature bits related to AVX.
 			if ( !bli_cpuid_has_features( eax, XGETBV_MASK_XMM |
 				                               XGETBV_MASK_YMM ) )
 			{
@@ -842,7 +842,7 @@ uint32_t bli_cpuid_query
 			// The OS can manage the state of 128-bit xmm (SSE) registers
 			// only if the xcr[1] bit is set. If it is not set, then
 			// clear all feature bits related to SSE (which means the
-			// entire bitfield is clear). 
+			// entire bitfield is clear).
 			if ( !bli_cpuid_has_features( eax, XGETBV_MASK_XMM ) )
 			{
 				*features = 0;
@@ -1025,6 +1025,7 @@ static uint32_t get_coretype
 {
 	int implementer = 0x00, part = 0x000;
 	*features = FEATURE_NEON;
+    bool has_sve = FALSE;
 
 #ifdef __linux__
 	if ( getauxval( AT_HWCAP ) & HWCAP_CPUID )
@@ -1033,7 +1034,7 @@ static uint32_t get_coretype
 		// /sys/devices/system/cpu/cpu0/regs/identification/midr_el1
 		// and split out in /proc/cpuinfo (with a tab before the colon):
 		// CPU part	: 0x0a1
-		
+
 		uint64_t midr_el1;
 		__asm("mrs %0, MIDR_EL1" : "=r" (midr_el1));
 		/*
@@ -1047,8 +1048,8 @@ static uint32_t get_coretype
 		implementer = (midr_el1 >> 24) & 0xFF;
 		part        = (midr_el1 >> 4)  & 0xFFF;
 	}
-	
-	bool has_sve = getauxval( AT_HWCAP ) & HWCAP_SVE;
+
+	has_sve = getauxval( AT_HWCAP ) & HWCAP_SVE;
 	if (has_sve)
 		*features |= FEATURE_SVE;
 #endif //__linux__
@@ -1097,7 +1098,7 @@ static uint32_t get_coretype
 	// CAVIUM_CPU_PART_THUNDERX2 0x0AF
 	// CAVIUM_CPU_PART_THUNDERX3 0x0B8  // taken from OpenBLAS
 	//
-	// BRCM_CPU_PART_BRAHMA_B53 0x100 
+	// BRCM_CPU_PART_BRAHMA_B53 0x100
 	// BRCM_CPU_PART_VULCAN 0x516
 	//
 	// QCOM_CPU_PART_FALKOR_V1 0x800
@@ -1210,7 +1211,7 @@ uint32_t bli_cpuid_query
 
 #elif defined(__arm__) || defined(_M_ARM) || defined(_ARCH_PPC)
 
-/* 
+/*
    I can't easily find documentation to do this as for aarch64, though
    it presumably could be unearthed from Linux code.  However, on
    Linux 5.2 (and Androids's 3.4), /proc/cpuinfo has this sort of

From d503703a728c8f77401d68054185229a24d814e8 Mon Sep 17 00:00:00 2001
From: Devin Matthews <damatthews@smu.edu>
Date: Wed, 26 Oct 2022 18:20:00 -0500
Subject: [PATCH 2/2] Add check to disable armsve on Apple M1.

---
 configure | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/configure b/configure
index a53f25380e..37399fbde2 100755
--- a/configure
+++ b/configure
@@ -1335,6 +1335,17 @@ blacklistbu_add()
 	fi
 }
 
+blacklistos_add()
+{
+	# Check whether we've already blacklisted the given sub-config so
+	# we don't output redundant messages.
+	if [ $(is_in_list "$1" "${config_blist}") == "false" ]; then
+
+		echowarn "The operating system does not support building '$1'; adding to blacklist."
+		config_blist="${config_blist} $1"
+	fi
+}
+
 blacklist_init()
 {
 	config_blist=""
@@ -1989,6 +2000,13 @@ check_assembler()
 	fi
 }
 
+check_os()
+{
+	if [[ "$(uname -s)" == "Darwin" && "$(uname -m)" == "arm64" ]]; then
+		blacklistos_add "armsve"
+	fi
+}
+
 try_assemble()
 {
 	local cc cflags asm_src asm_base asm_bin rval
@@ -2886,6 +2904,9 @@ main()
 	get_binutils_version
 	check_assembler
 
+    # Check if there is any incompatibility due to the operating system.
+	check_os
+
 	# Remove duplicates and whitespace from the blacklist.
 	blacklist_cleanup