[clang] [llvm] [AArch64] Remove Automatic Enablement of FEAT_F32MM (PR #85203)

Thu Mar 14 03:24:17 PDT 2024

https://github.com/Stylie777 created https://github.com/llvm/llvm-project/pull/85203

When `+sve` is passed in the command line, if the Architecture being targeted is V8.6A/V9.1A or later, `+f32mm` is also added. This enables FEAT_32MM, however at the time of writing no CPU's support this. This leads to the FEAT_32MM instructions being compiled for CPU's that do not support them.

This commit removes the automatic enablement, however the option is still able to be used by passing `+f32mm`.

>From b5a0ad5d4b7ab3510cd00d178da245761e34d32d Mon Sep 17 00:00:00 2001
From: Jack Styles <jack.styles at arm.com>
Date: Thu, 14 Mar 2024 10:06:05 +0000
Subject: [PATCH] [AArch64] Remove Automatic Enablement of FEAT_F32MM

When `+sve` is passed in the command line, if the Architecture being targeted is V8.6A/V9.1A or later, `+f32mm` is also added. This enables FEAT_32MM, however at the time of writing no CPU's support this. This leads to the FEAT_32MM instructions being compiled for CPU's that do not support them.

This commit removes the automatic enablement, however the option is still able to be used by passing `+f32mm`.
---
 clang/test/Driver/aarch64-sve.c                   | 9 ++++-----
 clang/test/Preprocessor/aarch64-target-features.c | 2 +-
 llvm/docs/ReleaseNotes.rst                        | 1 +
 llvm/lib/TargetParser/AArch64TargetParser.cpp     | 5 -----
 llvm/unittests/TargetParser/TargetParserTest.cpp  | 9 +--------
 5 files changed, 7 insertions(+), 19 deletions(-)

diff --git a/clang/test/Driver/aarch64-sve.c b/clang/test/Driver/aarch64-sve.c
index f34b2700deb91c..4a33c2e3c8d367 100644
--- a/clang/test/Driver/aarch64-sve.c
+++ b/clang/test/Driver/aarch64-sve.c
@@ -6,12 +6,11 @@
 // RUN: %clang --target=aarch64 -march=armv8.6a -### -c %s 2>&1 | FileCheck -check-prefix=GENERICV8A-NOSVE %s
 // GENERICV8A-NOSVE-NOT: "-target-feature" "+sve"
 
-// The 32-bit floating point matrix multiply extension is enabled by default
-// for armv8.6-a targets (or later) with SVE, and can optionally be enabled for
-// any target from armv8.2a onwards (we don't enforce not using it with earlier
-// targets).
+// The 32-bit floating point matrix multiply extension is an optional feature
+// that can be used for any target from armv8.2a and onwards. This can be
+// enabled using the `+f32mm` option.`.
 // RUN: %clang --target=aarch64 -march=armv8.6a       -### -c %s 2>&1 | FileCheck -check-prefix=NO-F32MM %s
-// RUN: %clang --target=aarch64 -march=armv8.6a+sve   -### -c %s 2>&1 | FileCheck -check-prefix=F32MM %s
+// RUN: %clang --target=aarch64 -march=armv8.6a+sve+f32mm   -### -c %s 2>&1 | FileCheck -check-prefix=F32MM %s
 // RUN: %clang --target=aarch64 -march=armv8.5a+f32mm -### -c %s 2>&1 | FileCheck -check-prefix=F32MM %s
 // NO-F32MM-NOT: "-target-feature" "+f32mm"
 // F32MM: "-target-feature" "+f32mm"
diff --git a/clang/test/Preprocessor/aarch64-target-features.c b/clang/test/Preprocessor/aarch64-target-features.c
index 6ec4dcd60cf601..ab06cc7983ae85 100644
--- a/clang/test/Preprocessor/aarch64-target-features.c
+++ b/clang/test/Preprocessor/aarch64-target-features.c
@@ -196,7 +196,7 @@
 // CHECK-8_6-NOT: __ARM_FEATURE_SHA3 1
 // CHECK-8_6-NOT: __ARM_FEATURE_SM4 1
 
-// RUN: %clang -target aarch64-none-linux-gnu -march=armv8.6-a+sve -x c -E -dM %s -o - | FileCheck --check-prefix=CHECK-SVE-8_6 %s
+// RUN: %clang -target aarch64-none-linux-gnu -march=armv8.6-a+sve+f32mm -x c -E -dM %s -o - | FileCheck --check-prefix=CHECK-SVE-8_6 %s
 // CHECK-SVE-8_6: __ARM_FEATURE_SVE 1
 // CHECK-SVE-8_6: __ARM_FEATURE_SVE_BF16 1
 // CHECK-SVE-8_6: __ARM_FEATURE_SVE_MATMUL_FP32 1
diff --git a/llvm/docs/ReleaseNotes.rst b/llvm/docs/ReleaseNotes.rst
index 7be51730663bd1..b65a108a9b6a36 100644
--- a/llvm/docs/ReleaseNotes.rst
+++ b/llvm/docs/ReleaseNotes.rst
@@ -74,6 +74,7 @@ Changes to the AMDGPU Backend
 
 Changes to the ARM Backend
 --------------------------
+* FEAT_F32MM is no longer activated by default when using `+sve` on v8.6-A or greater. The feature is still availble and can be using by adding `+f32mm` to the command line options.
 
 Changes to the AVR Backend
 --------------------------
diff --git a/llvm/lib/TargetParser/AArch64TargetParser.cpp b/llvm/lib/TargetParser/AArch64TargetParser.cpp
index e36832f563eed8..71099462d5ecff 100644
--- a/llvm/lib/TargetParser/AArch64TargetParser.cpp
+++ b/llvm/lib/TargetParser/AArch64TargetParser.cpp
@@ -186,11 +186,6 @@ void AArch64::ExtensionSet::enable(ArchExtKind E) {
   // Special cases for dependencies which vary depending on the base
   // architecture version.
   if (BaseArch) {
-    // +sve implies +f32mm if the base architecture is v8.6A+ or v9.1A+
-    // It isn't the case in general that sve implies both f64mm and f32mm
-    if (E == AEK_SVE && BaseArch->is_superset(ARMV8_6A))
-      enable(AEK_F32MM);
-
     // +fp16 implies +fp16fml for v8.4A+, but not v9.0-A+
     if (E == AEK_FP16 && BaseArch->is_superset(ARMV8_4A) &&
         !BaseArch->is_superset(ARMV9A))
diff --git a/llvm/unittests/TargetParser/TargetParserTest.cpp b/llvm/unittests/TargetParser/TargetParserTest.cpp
index 3773f59a3c5af9..a21846f5b4940b 100644
--- a/llvm/unittests/TargetParser/TargetParserTest.cpp
+++ b/llvm/unittests/TargetParser/TargetParserTest.cpp
@@ -2314,13 +2314,6 @@ AArch64ExtensionDependenciesBaseArchTestParams
          {},
          {"aes", "sha2", "sha3", "sm4"}},
 
-        // +sve implies +f32mm if the base architecture is v8.6A+ or v9.1A+, but
-        // not earlier architectures.
-        {AArch64::ARMV8_5A, {"sve"}, {"sve"}, {"f32mm"}},
-        {AArch64::ARMV9A, {"sve"}, {"sve"}, {"f32mm"}},
-        {AArch64::ARMV8_6A, {"sve"}, {"sve", "f32mm"}, {}},
-        {AArch64::ARMV9_1A, {"sve"}, {"sve", "f32mm"}, {}},
-
         // +fp16 implies +fp16fml for v8.4A+, but not v9.0-A+
         {AArch64::ARMV8_3A, {"fp16"}, {"fullfp16"}, {"fp16fml"}},
         {AArch64::ARMV9A, {"fp16"}, {"fullfp16"}, {"fp16fml"}},
@@ -2487,7 +2480,7 @@ AArch64ExtensionDependenciesBaseCPUTestParams
          {}},
         {"cortex-a520",
          {},
-         {"v9.2a",    "bf16",     "crc",     "dotprod", "f32mm",        "flagm",
+         {"v9.2a",    "bf16",     "crc",     "dotprod",        "flagm",
           "fp-armv8", "fullfp16", "fp16fml", "i8mm",    "lse",          "mte",
           "pauth",    "perfmon",  "predres", "ras",     "rcpc",         "rdm",
           "sb",       "neon",     "ssbs",    "sve",     "sve2-bitperm", "sve2"},

