[X86][Costmodel] `trunc v8i64 to v16i16/v32i16` can appear after legalization, cost is same as for `trunc v8i64 to v8i16`

[lebedev.ri]

Summary

Same as D113842, but for i64

[RKSimon]

LGTM

[lebedev.ri]

In D113843#3129765, @RKSimon wrote:

LGTM

Thank you for the review. I assume the i32 version (D113842) is also fine?

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

static const TypeConversionCostTblEntry AVX512DQConversionTbl[] = {
     { ISD::FP_TO_UINT,  MVT::v8i64,  MVT::v8f32,  1 },
     { ISD::FP_TO_UINT,  MVT::v8i64,  MVT::v8f64,  1 },
   };
 
   // TODO: For AVX512DQ + AVX512VL, we also have cheap casts for 128-bit and
   // 256-bit wide vectors.
 
   static const TypeConversionCostTblEntry AVX512FConversionTbl[] = {
     { ISD::FP_EXTEND, MVT::v8f64,   MVT::v8f32,  1 },
     { ISD::FP_EXTEND, MVT::v8f64,   MVT::v16f32, 3 },
     { ISD::FP_ROUND,  MVT::v8f32,   MVT::v8f64,  1 },
 
     { ISD::TRUNCATE,  MVT::v2i1,    MVT::v2i8,   3 }, // sext+vpslld+vptestmd
     { ISD::TRUNCATE,  MVT::v4i1,    MVT::v4i8,   3 }, // sext+vpslld+vptestmd
     { ISD::TRUNCATE,  MVT::v8i1,    MVT::v8i8,   3 }, // sext+vpslld+vptestmd
     { ISD::TRUNCATE,  MVT::v16i1,   MVT::v16i8,  3 }, // sext+vpslld+vptestmd
     { ISD::TRUNCATE,  MVT::v2i1,    MVT::v2i16,  3 }, // sext+vpsllq+vptestmq
static const TypeConversionCostTblEntry AVX512FConversionTbl[] = {
     { ISD::TRUNCATE,  MVT::v4i8,    MVT::v4i32,  2 }, // vpmovdb
     { ISD::TRUNCATE,  MVT::v16i8,   MVT::v16i32, 2 }, // vpmovdb
     { ISD::TRUNCATE,  MVT::v16i16,  MVT::v16i32, 2 }, // vpmovdw
     { ISD::TRUNCATE,  MVT::v32i16,  MVT::v16i32, 2 }, // vpmovdw
     { ISD::TRUNCATE,  MVT::v2i8,    MVT::v2i64,  2 }, // vpmovqb
     { ISD::TRUNCATE,  MVT::v2i16,   MVT::v2i64,  1 }, // vpshufb
     { ISD::TRUNCATE,  MVT::v8i8,    MVT::v8i64,  2 }, // vpmovqb
     { ISD::TRUNCATE,  MVT::v8i16,   MVT::v8i64,  2 }, // vpmovqw
+    { ISD::TRUNCATE,  MVT::v16i16,  MVT::v8i64,  2 }, // vpmovqw
+    { ISD::TRUNCATE,  MVT::v32i16,  MVT::v8i64,  2 }, // vpmovqw
     { ISD::TRUNCATE,  MVT::v8i32,   MVT::v8i64,  1 }, // vpmovqd
     { ISD::TRUNCATE,  MVT::v4i32,   MVT::v4i64,  1 }, // zmm vpmovqd
     { ISD::TRUNCATE,  MVT::v16i8,   MVT::v16i64, 5 },// 2*vpmovqd+concat+vpmovdb
 
     { ISD::TRUNCATE,  MVT::v16i8,  MVT::v16i16,  3 }, // extend to v16i32
     { ISD::TRUNCATE,  MVT::v32i8,  MVT::v32i16,  8 },
 
     // Sign extend is zmm vpternlogd+vptruncdb.

llvm/test/Analysis/CostModel/X86/trunc.ll

 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i64 = trunc i64 undef to i16
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = trunc <2 x i64> undef to <2 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V3i64 = trunc <3 x i64> undef to <3 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = trunc <4 x i64> undef to <4 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V5i64 = trunc <5 x i64> undef to <5 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 109 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 226 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 276 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 318 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 460 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 552 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 644 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 920 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1104 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1288 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V3i32 = trunc <3 x i32> undef to <3 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = trunc <4 x i32> undef to <4 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V5i32 = trunc <5 x i32> undef to <5 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6i32 = trunc <6 x i32> undef to <6 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7i32 = trunc <7 x i32> undef to <7 x i16>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>