[AArch64] Mark aarch64_set_fpmr as IntrWriteMem

llvm.aarch64.set.fpmr only writes to inaccessible memory.
Tag it with the IntrWriteMem and IntrInaccessibleMemOnly properties so the
optimiser can treat it as a pure write.

The original patch did not add this property, causing the intrinsic
to be conservatively treated as readwrite. This commit fixes that.

Author: CarolineConcatto

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_cvt.c

@@ -330,3 +330,6 @@ mfloat8x8_t test_vcvt_mf8_f16_fpm(float16x4_t vn, float16x4_t vm, fpm_t fpm) {
 mfloat8x16_t test_vcvtq_mf8_f16_fpm(float16x8_t vn, float16x8_t vm, fpm_t fpm) {
   return vcvtq_mf8_f16_fpm(vn, vm, fpm);
 }
+
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_fdot.c

@@ -83,6 +83,8 @@ float16x4_t test_vdot_lane_f16(float16x4_t vd, mfloat8x8_t vn, mfloat8x8_t vm, f
   return vdot_lane_f16_mf8_fpm(vd, vn, vm, 3, fpmr);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <4 x half> @test_vdot_laneq_f16(
 // CHECK-SAME: <4 x half> noundef [[VD:%.*]], <8 x i8> [[VN:%.*]], <16 x i8> [[VM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -268,3 +270,5 @@ float32x4_t test_vdotq_lane_f32(float32x4_t vd, mfloat8x16_t vn, mfloat8x8_t vm,
 float32x4_t test_vdotq_laneq_f32(float32x4_t vd, mfloat8x16_t vn, mfloat8x16_t vm, fpm_t fpmr) {
   return vdotq_laneq_f32_mf8_fpm(vd, vn, vm, 3, fpmr);
 }
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_neon_fp8_fmla.c

@@ -154,6 +154,8 @@ float16x8_t test_vmlalb_lane(float16x8_t vd, mfloat8x16_t vn, mfloat8x8_t vm, fp
   return vmlalbq_lane_f16_mf8_fpm(vd, vn, vm, 0, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <8 x half> @test_vmlalb_laneq(
 // CHECK-SAME: <8 x half> noundef [[VD:%.*]], <16 x i8> [[VN:%.*]], <16 x i8> [[VM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -379,3 +381,5 @@ float32x4_t test_vmlalltt_lane(float32x4_t vd, mfloat8x16_t vn, mfloat8x8_t vm,
 float32x4_t test_vmlalltt_laneq(float32x4_t vd, mfloat8x16_t vn, mfloat8x16_t vm, fpm_t fpm) {
   return vmlallttq_laneq_f32_mf8_fpm(vd, vn, vm, 15, fpm);
 }
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_cvt.c

@@ -32,6 +32,8 @@ svmfloat8_t test_cvt_f16_x2(svfloat16x2_t zn, fpm_t fpmr)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvt_mf8,_f16_x2,_fpm)(zn, fpmr);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: @test_cvt_f32_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR:%.*]])
@@ -207,3 +209,5 @@ svbfloat16x2_t test_cvtl1_bf16_x2(svmfloat8_t zn, fpm_t fpmr)  __arm_streaming {
 svbfloat16x2_t test_cvtl2_bf16_x2(svmfloat8_t zn, fpm_t fpmr)  __arm_streaming {
   return SVE_ACLE_FUNC(svcvtl2_bf16,_mf8,_x2_fpm)(zn, fpmr);
 }
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_fmopa.c

@@ -35,6 +35,8 @@ void test_svmopa_za16_mf8_m(svbool_t pn, svbool_t pm, svmfloat8_t zn,
   SVE_ACLE_FUNC(svmopa_za16,_mf8,_m_fpm)(1, pn, pm, zn, zm, fpmr);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+//
 // CHECK-LABEL: define dso_local void @test_svmopa_za32_mf8_m(
 // CHECK-SAME: <vscale x 16 x i1> [[PN:%.*]], <vscale x 16 x i1> [[PM:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -53,3 +55,6 @@ void test_svmopa_za32_mf8_m(svbool_t pn, svbool_t pm, svmfloat8_t zn,
                             svmfloat8_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {
   SVE_ACLE_FUNC(svmopa_za32,_mf8,_m_fpm)(3, pn, pm, zn, zm, fpmr);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_mla.c

@@ -35,6 +35,8 @@ void test_svmla_lane_za16_vg2x1(uint32_t slice, svmfloat8_t zn, svmfloat8_t zm,
     SME_ACLE_FUNC(svmla_lane_za16,_mf8,_vg2x1_fpm,,)(slice, zn, zm, 0, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local void @test_svmla_lane_za16_vg2x2(
 // CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -314,3 +316,6 @@ void test_svmla_multi_za32_vg4x2(uint32_t slice, svmfloat8x2_t zn, svmfloat8x2_t
 void test_svmla_multi_za32_vg4x4(uint32_t slice, svmfloat8x4_t zn, svmfloat8x4_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
     SME_ACLE_FUNC(svmla_za32,_mf8,_vg4x4_fpm,,)(slice, zn, zm, fpm);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sve2_fp8_cvt.c

@@ -46,6 +46,8 @@ svbfloat16_t test_svcvt1_bf16_mf8(svmfloat8_t zn, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svcvt1_bf16,_mf8,_fpm)(zn, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <vscale x 8 x bfloat> @test_svcvt2_bf16_mf8(
 // CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -171,3 +173,6 @@ svfloat16_t test_svcvtlt1_f16_mf8(svmfloat8_t zn, fpm_t fpm) STREAMING {
 svfloat16_t test_svcvtlt2_f16_mf8(svmfloat8_t zn, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svcvtlt2_f16,_mf8,_fpm)(zn, fpm);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sve2_fp8_cvtn.c

@@ -46,6 +46,8 @@ svmfloat8_t test_svcvtn_f8_bf16(svbfloat16x2_t zn_zm, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svcvtn_mf8,_bf16_x2,_fpm)(zn_zm, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <vscale x 16 x i8> @test_svcvtn_f8_f16(
 // CHECK-SAME: <vscale x 8 x half> [[ZN_ZM_COERCE0:%.*]], <vscale x 8 x half> [[ZN_ZM_COERCE1:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -99,3 +101,6 @@ svmfloat8_t test_svcvtnb_f8_f32(svfloat32x2_t zn_zm, fpm_t fpm) STREAMING {
 svmfloat8_t test_svcvtnt_f8_f32(svmfloat8_t zd, svfloat32x2_t zn_zm, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svcvtnt_mf8,_f32_x2,_fpm)(zd, zn_zm, fpm);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sve2_fp8_fdot.c

@@ -46,6 +46,8 @@ svfloat32_t test_svdot_f32_mf8(svfloat32_t zda, svmfloat8_t zn, svmfloat8_t zm,
   return SVE_ACLE_FUNC(svdot,_f32_mf8,_fpm)(zda, zn, zm, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64)  [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <vscale x 4 x float> @test_svdot_n_f32_mf8(
 // CHECK-SAME: <vscale x 4 x float> [[ZDA:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <1 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -147,3 +149,6 @@ svfloat32_t test_svdot_lane_f32_mf8(svfloat32_t zda, svmfloat8_t zn, svmfloat8_t
 svfloat16_t test_svdot_lane_f16_mf8(svfloat16_t zda, svmfloat8_t zn, svmfloat8_t zm, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svdot_lane,_f16_mf8,_fpm)(zda, zn, zm, 7, fpm);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }

clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sve2_fp8_fmla.c

@@ -46,6 +46,8 @@ svfloat16_t test_svmlalb_f16_mf8(svfloat16_t zda, svmfloat8_t zn, svmfloat8_t zm
   return SVE_ACLE_FUNC(svmlalb,_f16_mf8,_fpm)(zda, zn, zm, fpm);
 }
 
+// CHECK: declare void @llvm.aarch64.set.fpmr(i64) [[ATTR1:#.*]]
+
 // CHECK-LABEL: define dso_local <vscale x 8 x half> @test_svmlalb_n_f16_mf8(
 // CHECK-SAME: <vscale x 8 x half> [[ZDA:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <1 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
 // CHECK-NEXT:  [[ENTRY:.*:]]
@@ -387,3 +389,6 @@ svfloat32_t test_svmlalltb_lane_f32_mf8(svfloat32_t zda, svmfloat8_t zn, svmfloa
 svfloat32_t test_svmlalltt_lane_f32_mf8(svfloat32_t zda, svmfloat8_t zn, svmfloat8_t zm, fpm_t fpm) STREAMING {
   return SVE_ACLE_FUNC(svmlalltt_lane,_f32_mf8,_fpm)(zda, zn, zm, 7, fpm);
 }
+
+
+// CHECK: attributes [[ATTR1]] = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }