[RISCV][CostModel] Add test for fptoi_sat

[Miss_Grape]

The costmodel setting of llvm.fptoui.sat is incorrect， need to do.

[reames]

LGTM

Please rename add the scalable vector types as well.

You may find https://github.com/preames/riscv-llvm-cost-model-gen/ useful as a starting point for generating these.

llvm/test/Analysis/CostModel/RISCV/fptoi_sat.ll

+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=riscv64 -mattr=+v,+f -riscv-v-vector-bits-min=256 -passes="print<cost-model>" -disable-output | FileCheck %s
+; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -S -mtriple=riscv64 -mattr=+v,+f | FileCheck %s
+
+define void @casts() {
+; CHECK-LABEL: 'casts'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+  %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+  %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+  %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+  %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+  %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+  %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+  %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+  %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+  %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+
+  %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+  %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+  %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+  %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+  %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+  %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+  %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+  %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+  %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+  %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+
+  %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+  %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+  %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+  %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+  %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+  %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+  %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+  %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+  %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+  %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+
+  %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+  %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+  %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+  %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+  %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+  %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+  %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+  %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+  %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+  %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+
+  %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+  %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+  %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+  %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+  %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+  %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+  %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+  %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+  %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+  %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+
+  %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+  %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+  %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+  %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+  %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+  %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+  %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+  %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+  %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+  %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+
+  %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+  %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+  %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+  %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+  %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+  %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+  %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+  %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+  %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+  %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+
+  %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+  %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+  %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+  %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+  %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+  %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+  %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+  %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+  %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+  %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+
+  %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+  %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+  %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+  %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+  %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+  %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+  %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+  %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+  %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+  %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+
+  %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+  %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+  %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+  %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+  %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+  %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+  %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+  %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+  %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+  %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+
+  ret void
+}
+
+
+declare i1 @llvm.fptosi.sat.i1.f32(float)
+declare i1 @llvm.fptoui.sat.i1.f32(float)
+declare i8 @llvm.fptosi.sat.i8.f32(float)
+declare i8 @llvm.fptoui.sat.i8.f32(float)
+declare i16 @llvm.fptosi.sat.i16.f32(float)
+declare i16 @llvm.fptoui.sat.i16.f32(float)
+declare i32 @llvm.fptosi.sat.i32.f32(float)
+declare i32 @llvm.fptoui.sat.i32.f32(float)
+declare i64 @llvm.fptosi.sat.i64.f32(float)
+declare i64 @llvm.fptoui.sat.i64.f32(float)
+
+declare i1 @llvm.fptosi.sat.i1.f64(double)
+declare i1 @llvm.fptoui.sat.i1.f64(double)
+declare i8 @llvm.fptosi.sat.i8.f64(double)
+declare i8 @llvm.fptoui.sat.i8.f64(double)
+declare i16 @llvm.fptosi.sat.i16.f64(double)
+declare i16 @llvm.fptoui.sat.i16.f64(double)
+declare i32 @llvm.fptosi.sat.i32.f64(double)
+declare i32 @llvm.fptoui.sat.i32.f64(double)
+declare i64 @llvm.fptosi.sat.i64.f64(double)
+declare i64 @llvm.fptoui.sat.i64.f64(double)
+
+declare <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float>)
+declare <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float>)
+declare <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float>)
+declare <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float>)
+declare <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float>)
+declare <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float>)
+declare <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float>)
+declare <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float>)
+declare <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float>)
+declare <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float>)
+
+declare <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double>)
+declare <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double>)
+declare <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double>)
+declare <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double>)
+declare <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double>)
+declare <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double>)
+declare <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double>)
+declare <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double>)
+declare <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double>)
+declare <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double>)
+
+declare <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float>)
+declare <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float>)
+declare <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float>)
+declare <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float>)
+declare <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float>)
+declare <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float>)
+declare <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float>)
+declare <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float>)
+declare <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float>)
+declare <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float>)
+
+declare <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double>)
+declare <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double>)
+declare <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double>)
+declare <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double>)
+declare <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double>)
+declare <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double>)
+declare <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double>)
+declare <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double>)
+declare <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double>)
+declare <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double>)
+
+declare <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float>)
+declare <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float>)
+declare <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float>)
+declare <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float>)
+declare <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float>)
+declare <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float>)
+declare <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float>)
+declare <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float>)
+declare <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float>)
+declare <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float>)
+
+declare <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double>)
+declare <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double>)
+declare <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double>)
+declare <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double>)
+declare <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double>)
+declare <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double>)
+declare <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double>)
+declare <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double>)
+declare <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double>)
+declare <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double>)
+
+declare <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float>)
+declare <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float>)
+declare <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float>)
+declare <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float>)
+declare <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float>)
+declare <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float>)
+declare <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float>)
+declare <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float>)
+declare <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float>)
+declare <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float>)
+
+declare <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double>)
+declare <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double>)
+declare <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double>)
+declare <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double>)
+declare <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double>)
+declare <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double>)
+declare <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double>)
+declare <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double>)
+declare <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double>)
+declare <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double>)
+