[clang] [CIR][AArch64] Handle constrained Neon FMA and sqrt (PR #218307)

[Yair Ben Avraham via cfe-commits]

================
@@ -8,6 +8,24 @@
 // RUN: -flax-vector-conversions=none -disable-O0-optnone -emit-llvm -o - %s \
 // RUN: | opt -S -passes=mem2reg,sroa \
 // RUN: | FileCheck --check-prefix=CONSTRAINED --implicit-check-not=fpexcept.maytrap %s
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a \
+// RUN: -fexperimental-strict-floating-point -ffp-exception-behavior=maytrap -DEXCEPT=1 \
+// RUN: -flax-vector-conversions=none -disable-O0-optnone -fclangir -emit-cir -o - %s \
+// RUN: | FileCheck --check-prefix=CIR --implicit-check-not='cir.call_llvm_intrinsic "fma"' --implicit-check-not='cir.call_llvm_intrinsic "sqrt"' %s %}
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a \
+// RUN: -fexperimental-strict-floating-point -ffp-exception-behavior=maytrap -DEXCEPT=1 \
+// RUN: -flax-vector-conversions=none -disable-O0-optnone -fclangir -emit-llvm -o - %s \
+// RUN: | opt -S -passes=mem2reg,sroa \
+// RUN: | FileCheck --check-prefix=LLVM --implicit-check-not=fpexcept.maytrap --implicit-check-not=' @llvm.fma.' --implicit-check-not=' @llvm.sqrt.' %s %}
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a \
+// RUN: -fexperimental-strict-floating-point -ffp-exception-behavior=strict \
+// RUN: -flax-vector-conversions=none -disable-O0-optnone -fclangir -emit-cir -o - %s \
+// RUN: | FileCheck --check-prefix=CIR --implicit-check-not='cir.call_llvm_intrinsic "fma"' --implicit-check-not='cir.call_llvm_intrinsic "sqrt"' %s %}
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -target-feature +fullfp16 -target-feature +v8.2a \
+// RUN: -fexperimental-strict-floating-point -ffp-exception-behavior=strict \
+// RUN: -flax-vector-conversions=none -disable-O0-optnone -fclangir -emit-llvm -o - %s \
+// RUN: | opt -S -passes=mem2reg,sroa \
+// RUN: | FileCheck --check-prefix=LLVM --implicit-check-not=' @llvm.fma.' --implicit-check-not=' @llvm.sqrt.' %s %}
----------------
yairbenavraham wrote:

I moved the new checks into neon/*-constrained.c files. The old test file is unchanged in the final diff.

https://github.com/llvm/llvm-project/pull/218307