[RISCV] Add cost model for fp-mask cast op.

[jacquesguan]

Summary

The cost of convert from or to mask vector is different from other cases. We could not use PowDiff to calculate it. This patch set it to 3 as we use 3 instruction to make it.

[reames]

LGTM w/required change before landing

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

118–122

You have a bunch of white space changes here which make the test diffs impossible to read.

Please update the test diffs once on a clean checkout, commit the change, and then apply your change and regenerate the test diffs.

[jacquesguan]

rebase main.

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

118–122

Done.

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

if (isa<VectorType>(Dst) && isa<VectorType>(Src)) {
     case ISD::FP_EXTEND:
     case ISD::FP_ROUND:
       // Counts of narrow/widen instructions.
       return std::abs(PowDiff);
     case ISD::FP_TO_SINT:
     case ISD::FP_TO_UINT:
     case ISD::SINT_TO_FP:
     case ISD::UINT_TO_FP:
+      if (Src->getScalarSizeInBits() == 1 || Dst->getScalarSizeInBits() == 1) {
+        // The cost of convert from or to mask vector is different from other
+        // cases. We could not use PowDiff to calculate it.
+        // For mask vector to fp, we should use the following instructions:
+        // vmv.v.i v8, 0
+        // vmerge.vim v8, v8, -1, v0
+        // vfcvt.f.x.v v8, v8
+
+        // And for fp vector to mask, we use:
+        // vfncvt.rtz.x.f.w v9, v8
+        // vand.vi v8, v9, 1
+        // vmsne.vi v0, v8, 0
+        return 3;
+      }
       if (std::abs(PowDiff) <= 1)
         return 1;
       // Backend could lower (v[sz]ext i8 to double) to vfcvt(v[sz]ext.f8 i8),
       // so it only need two conversion.
       if (Src->isIntOrIntVectorTy())
         return 2;
       // Counts of narrow/widen instructions.
       return std::abs(PowDiff);

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

 define void @fpext() {
 ; RISCV32-LABEL: 'fpext'
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v0 = fpext <8 x half> undef to <8 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v1 = fpext <8 x half> undef to <8 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2 = fpext <8 x float> undef to <8 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; RISCV64-LABEL: 'fpext'
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v0 = fpext <8 x half> undef to <8 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v1 = fpext <8 x half> undef to <8 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2 = fpext <8 x float> undef to <8 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %v0 = fpext <8 x half>  undef to <8 x float>
   %v1 = fpext <8 x half>  undef to <8 x double>
   %v2 = fpext <8 x float> undef to <8 x double>
   ret void
 }
 
 define void @ftrunc() {
 ; RISCV32-LABEL: 'ftrunc'

 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s4 = fptosi <4 x float> undef to <4 x i16>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s5 = fptosi <4 x double> undef to <4 x i16>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s6 = fptosi <4 x half> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = fptosi <4 x float> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = fptosi <4 x double> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s9 = fptosi <4 x half> undef to <4 x i64>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = fptosi <4 x float> undef to <4 x i64>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s11 = fptosi <4 x double> undef to <4 x i16>
+; RISCV32-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s12 = fptosi <4 x double> undef to <4 x i1>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u0 = fptoui <4 x half> undef to <4 x i8>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u1 = fptoui <4 x float> undef to <4 x i8>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u2 = fptoui <4 x double> undef to <4 x i8>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u3 = fptoui <4 x half> undef to <4 x i16>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u4 = fptoui <4 x float> undef to <4 x i16>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u5 = fptoui <4 x double> undef to <4 x i16>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u6 = fptoui <4 x half> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u7 = fptoui <4 x float> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u8 = fptoui <4 x double> undef to <4 x i32>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u9 = fptoui <4 x half> undef to <4 x i64>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u10 = fptoui <4 x float> undef to <4 x i64>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u11 = fptoui <4 x double> undef to <4 x i16>
+; RISCV32-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u12 = fptoui <4 x double> undef to <4 x i1>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; RISCV64-LABEL: 'fp_to_int'
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s0 = fptosi <4 x half> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s1 = fptosi <4 x float> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s2 = fptosi <4 x double> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s3 = fptosi <4 x half> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s4 = fptosi <4 x float> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s5 = fptosi <4 x double> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s6 = fptosi <4 x half> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = fptosi <4 x float> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = fptosi <4 x double> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s9 = fptosi <4 x half> undef to <4 x i64>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = fptosi <4 x float> undef to <4 x i64>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s11 = fptosi <4 x double> undef to <4 x i16>
+; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s12 = fptosi <4 x double> undef to <4 x i1>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u0 = fptoui <4 x half> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u1 = fptoui <4 x float> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u2 = fptoui <4 x double> undef to <4 x i8>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u3 = fptoui <4 x half> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u4 = fptoui <4 x float> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u5 = fptoui <4 x double> undef to <4 x i16>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u6 = fptoui <4 x half> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u7 = fptoui <4 x float> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u8 = fptoui <4 x double> undef to <4 x i32>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u9 = fptoui <4 x half> undef to <4 x i64>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u10 = fptoui <4 x float> undef to <4 x i64>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u11 = fptoui <4 x double> undef to <4 x i16>
+; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u12 = fptoui <4 x double> undef to <4 x i1>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %s0 = fptosi <4 x half> undef to <4 x i8>
   %s1 = fptosi <4 x float> undef to <4 x i8>
   %s2 = fptosi <4 x double> undef to <4 x i8>
   %s3 = fptosi <4 x half> undef to <4 x i16>
   %s4 = fptosi <4 x float> undef to <4 x i16>
   %s5 = fptosi <4 x double> undef to <4 x i16>
   %s6 = fptosi <4 x half> undef to <4 x i32>
   %s7 = fptosi <4 x float> undef to <4 x i32>
   %s8 = fptosi <4 x double> undef to <4 x i32>
   %s9 = fptosi <4 x half> undef to <4 x i64>
   %s10 = fptosi <4 x float> undef to <4 x i64>
   %s11 = fptosi <4 x double> undef to <4 x i16>
+  %s12 = fptosi <4 x double> undef to <4 x i1>
 
   %u0 = fptoui <4 x half> undef to <4 x i8>
   %u1 = fptoui <4 x float> undef to <4 x i8>
   %u2 = fptoui <4 x double> undef to <4 x i8>
   %u3 = fptoui <4 x half> undef to <4 x i16>
   %u4 = fptoui <4 x float> undef to <4 x i16>
   %u5 = fptoui <4 x double> undef to <4 x i16>
   %u6 = fptoui <4 x half> undef to <4 x i32>
   %u7 = fptoui <4 x float> undef to <4 x i32>
   %u8 = fptoui <4 x double> undef to <4 x i32>
   %u9 = fptoui <4 x half> undef to <4 x i64>
   %u10 = fptoui <4 x float> undef to <4 x i64>
   %u11 = fptoui <4 x double> undef to <4 x i16>
+  %u12 = fptoui <4 x double> undef to <4 x i1>
+
   ret void
 }
 
 define void @int_to_fp() {
 ; RISCV32-LABEL: 'int_to_fp'
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s0 = sitofp <4 x i8> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s1 = sitofp <4 x i8> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s2 = sitofp <4 x i8> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s3 = sitofp <4 x i16> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s4 = sitofp <4 x i16> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s5 = sitofp <4 x i16> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s6 = sitofp <4 x i32> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = sitofp <4 x i32> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = sitofp <4 x i32> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s9 = sitofp <4 x i64> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = sitofp <4 x i64> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s11 = sitofp <4 x i16> undef to <4 x double>
+; RISCV32-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s12 = sitofp <4 x i1> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u0 = uitofp <4 x i8> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u1 = uitofp <4 x i8> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u2 = uitofp <4 x i8> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u3 = uitofp <4 x i16> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u4 = uitofp <4 x i16> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u5 = uitofp <4 x i16> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u6 = uitofp <4 x i32> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u7 = uitofp <4 x i32> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u8 = uitofp <4 x i32> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u9 = uitofp <4 x i64> undef to <4 x half>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u10 = uitofp <4 x i64> undef to <4 x float>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u11 = uitofp <4 x i16> undef to <4 x double>
+; RISCV32-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u12 = uitofp <4 x i1> undef to <4 x double>
 ; RISCV32-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; RISCV64-LABEL: 'int_to_fp'
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s0 = sitofp <4 x i8> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s1 = sitofp <4 x i8> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s2 = sitofp <4 x i8> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s3 = sitofp <4 x i16> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s4 = sitofp <4 x i16> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s5 = sitofp <4 x i16> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s6 = sitofp <4 x i32> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = sitofp <4 x i32> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = sitofp <4 x i32> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s9 = sitofp <4 x i64> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = sitofp <4 x i64> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s11 = sitofp <4 x i16> undef to <4 x double>
+; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s12 = sitofp <4 x i1> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u0 = uitofp <4 x i8> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u1 = uitofp <4 x i8> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u2 = uitofp <4 x i8> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u3 = uitofp <4 x i16> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u4 = uitofp <4 x i16> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u5 = uitofp <4 x i16> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u6 = uitofp <4 x i32> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u7 = uitofp <4 x i32> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u8 = uitofp <4 x i32> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u9 = uitofp <4 x i64> undef to <4 x half>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %u10 = uitofp <4 x i64> undef to <4 x float>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %u11 = uitofp <4 x i16> undef to <4 x double>
+; RISCV64-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %u12 = uitofp <4 x i1> undef to <4 x double>
 ; RISCV64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %s0 = sitofp <4 x i8> undef to <4 x half>
   %s1 = sitofp <4 x i8> undef to <4 x float>
   %s2 = sitofp <4 x i8> undef to <4 x double>
   %s3 = sitofp <4 x i16> undef to <4 x half>
   %s4 = sitofp <4 x i16> undef to <4 x float>
   %s5 = sitofp <4 x i16> undef to <4 x double>
   %s6 = sitofp <4 x i32> undef to <4 x half>
   %s7 = sitofp <4 x i32> undef to <4 x float>
   %s8 = sitofp <4 x i32> undef to <4 x double>
   %s9 = sitofp <4 x i64> undef to <4 x half>
   %s10 = sitofp <4 x i64> undef to <4 x float>
   %s11 = sitofp <4 x i16> undef to <4 x double>
+  %s12 = sitofp <4 x i1> undef to <4 x double>
 
   %u0 = uitofp <4 x i8> undef to <4 x half>
   %u1 = uitofp <4 x i8> undef to <4 x float>
   %u2 = uitofp <4 x i8> undef to <4 x double>
   %u3 = uitofp <4 x i16> undef to <4 x half>
   %u4 = uitofp <4 x i16> undef to <4 x float>
   %u5 = uitofp <4 x i16> undef to <4 x double>
   %u6 = uitofp <4 x i32> undef to <4 x half>
   %u7 = uitofp <4 x i32> undef to <4 x float>
   %u8 = uitofp <4 x i32> undef to <4 x double>
   %u9 = uitofp <4 x i64> undef to <4 x half>
   %u10 = uitofp <4 x i64> undef to <4 x float>
   %u11 = uitofp <4 x i16> undef to <4 x double>
+  %u12 = uitofp <4 x i1> undef to <4 x double>
+
   ret void
 }