[Mlir-commits] [mlir] [mlir][llvm] Add more constrained FP operations (PR #213745)

[Tobias Gysi]

================
@@ -1216,6 +1218,87 @@ define void @experimental_constrained_fpext(float %s, <4 x float> %v) {
   ret void
 }
 
+; CHECK-LABEL: experimental_constrained_nearbyint
+define void @experimental_constrained_nearbyint(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.nearbyint %{{.*}} tonearest strict : f32
+  %1 = call float @llvm.experimental.constrained.nearbyint.f32(float %s, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.nearbyint %{{.*}} towardzero ignore : vector<4xf32>
+  %2 = call <4 x float> @llvm.experimental.constrained.nearbyint.v4f32(<4 x float> %v, metadata !"round.towardzero", metadata !"fpexcept.ignore")
+  ; CHECK: llvm.intr.experimental.constrained.nearbyint %{{.*}} tonearest strict {fastmathFlags = #llvm.fastmath<nsz>} : f32
+  %3 = call nsz float @llvm.experimental.constrained.nearbyint.f32(float %s, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_ceil
+define void @experimental_constrained_ceil(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.ceil %{{.*}} strict : f32
+  %1 = call float @llvm.experimental.constrained.ceil.f32(float %s, metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.ceil %{{.*}} ignore : vector<4xf32>
+  %2 = call <4 x float> @llvm.experimental.constrained.ceil.v4f32(<4 x float> %v, metadata !"fpexcept.ignore")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_maxnum
+define void @experimental_constrained_maxnum(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.maxnum %{{.*}}, %{{.*}} strict : f32
+  %1 = call float @llvm.experimental.constrained.maxnum.f32(float %s, float %s, metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.maxnum %{{.*}}, %{{.*}} ignore : vector<4xf32>
+  %2 = call <4 x float> @llvm.experimental.constrained.maxnum.v4f32(<4 x float> %v, <4 x float> %v, metadata !"fpexcept.ignore")
+  ; CHECK: llvm.intr.experimental.constrained.maxnum %{{.*}}, %{{.*}} strict {fastmathFlags = #llvm.fastmath<nsz>} : f32
+  %3 = call nsz float @llvm.experimental.constrained.maxnum.f32(float %s, float %s, metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_minnum
+define void @experimental_constrained_minnum(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.minnum %{{.*}}, %{{.*}} strict : f32
+  %1 = call float @llvm.experimental.constrained.minnum.f32(float %s, float %s, metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.minnum %{{.*}}, %{{.*}} ignore {fastmathFlags = #llvm.fastmath<nsz>} : f32
+  %2 = call nsz float @llvm.experimental.constrained.minnum.f32(float %s, float %s, metadata !"fpexcept.ignore")
+  ; CHECK: llvm.intr.experimental.constrained.minnum %{{.*}}, %{{.*}} strict {fastmathFlags = #llvm.fastmath<nsz>} : vector<4xf32>
+  %3 = call nsz <4 x float> @llvm.experimental.constrained.minnum.v4f32(<4 x float> %v, <4 x float> %v, metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_ldexp
+define void @experimental_constrained_ldexp(float %s, i32 %e) {
+  ; CHECK: llvm.intr.experimental.constrained.ldexp %{{.*}}, %{{.*}} tonearest strict : (f32, i32) -> f32
+  %1 = call float @llvm.experimental.constrained.ldexp.f32.i32(float %s, i32 %e, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_lrint
+define void @experimental_constrained_lrint(float %s) {
+  ; CHECK: llvm.intr.experimental.constrained.lrint %{{.*}} tonearest strict : f32 to i32
+  %1 = call i32 @llvm.experimental.constrained.lrint.i32.f32(float %s, metadata !"round.tonearest", metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_lround
+define void @experimental_constrained_lround(float %s) {
+  ; CHECK: llvm.intr.experimental.constrained.lround %{{.*}} strict : f32 to i32
+  %1 = call i32 @llvm.experimental.constrained.lround.i32.f32(float %s, metadata !"fpexcept.strict")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_fptosi
+define void @experimental_constrained_fptosi(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.fptosi %{{.*}} strict : f32 to i32
+  %1 = call i32 @llvm.experimental.constrained.fptosi.i32.f32(float %s, metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.fptosi %{{.*}} ignore : vector<4xf32> to vector<4xi32>
+  %2 = call <4 x i32> @llvm.experimental.constrained.fptosi.v4i32.v4f32(<4 x float> %v, metadata !"fpexcept.ignore")
+  ret void
+}
+
+; CHECK-LABEL: experimental_constrained_fcmp
+define void @experimental_constrained_fcmp(float %s, <4 x float> %v) {
+  ; CHECK: llvm.intr.experimental.constrained.fcmp oeq %{{.*}}, %{{.*}} strict : f32
+  %1 = call i1 @llvm.experimental.constrained.fcmp.f32(float %s, float %s, metadata !"oeq", metadata !"fpexcept.strict")
+  ; CHECK: llvm.intr.experimental.constrained.fcmps olt %{{.*}}, %{{.*}} ignore : vector<4xf32>
+  %2 = call <4 x i1> @llvm.experimental.constrained.fcmps.v4f32(<4 x float> %v, <4 x float> %v, metadata !"olt", metadata !"fpexcept.ignore")
+  ret void
+}
----------------
gysit wrote:

We should probably have an import test for every operation category (i.e. for every base class). It seems line some are missing such as the power intrinsic?

https://github.com/llvm/llvm-project/pull/213745