[CostModel][X86] Added tests for sdiv/udiv costs for uniform const and uniform const...

Shows poor costings in AVX1/AVX512BW for certain vector types

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@284748 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/Analysis/CostModel/X86/div.ll b/test/Analysis/CostModel/X86/div.ll
index 2fe58e19cf2b60733ed41b803249bcd3a0c5be5c..620420c24b28998586ba24364b015ed702efa328 100644 (file)
--- a/test/Analysis/CostModel/X86/div.ll
+++ b/test/Analysis/CostModel/X86/div.ll
@@ -114,3 +114,267 @@ define i32 @udiv() {
 
   ret i32 undef
 }\r
+\r
+; CHECK-LABEL: 'sdiv_uniformconst'
+define i32 @sdiv_uniformconst() {
+  ; CHECK: cost of 1 {{.*}} %I64 = sdiv
+  %I64 = sdiv i64 undef, 7
+  ; SSE: cost of 40 {{.*}} %V2i64 = sdiv
+  ; AVX: cost of 40 {{.*}} %V2i64 = sdiv
+  %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+  ; SSE: cost of 80 {{.*}} %V4i64 = sdiv
+  ; AVX: cost of 80 {{.*}} %V4i64 = sdiv
+  %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+  ; SSE: cost of 160 {{.*}} %V8i64 = sdiv
+  ; AVX: cost of 160 {{.*}} %V8i64 = sdiv
+  %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+
+  ; CHECK: cost of 1 {{.*}} %I32 = sdiv
+  %I32 = sdiv i32 undef, 7
+  ; SSE2: cost of 19 {{.*}} %V4i32 = sdiv
+  ; SSSE3: cost of 19 {{.*}} %V4i32 = sdiv
+  ; SSE42: cost of 15 {{.*}} %V4i32 = sdiv
+  ; AVX: cost of 15 {{.*}} %V4i32 = sdiv
+  %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+  ; SSE2: cost of 38 {{.*}} %V8i32 = sdiv
+  ; SSSE3: cost of 38 {{.*}} %V8i32 = sdiv
+  ; SSE42: cost of 30 {{.*}} %V8i32 = sdiv
+  ; AVX1: cost of 160 {{.*}} %V8i32 = sdiv
+  ; AVX2: cost of 15 {{.*}} %V8i32 = sdiv
+  ; AVX512: cost of 15 {{.*}} %V8i32 = sdiv
+  %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+  ; SSE2: cost of 76 {{.*}} %V16i32 = sdiv
+  ; SSSE3: cost of 76 {{.*}} %V16i32 = sdiv
+  ; SSE42: cost of 60 {{.*}} %V16i32 = sdiv
+  ; AVX1: cost of 320 {{.*}} %V16i32 = sdiv
+  ; AVX2: cost of 30 {{.*}} %V16i32 = sdiv
+  ; AVX512F: cost of 48 {{.*}} %V16i32 = sdiv
+  ; AVX512BW: cost of 320 {{.*}} %V16i32 = sdiv
+  %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+
+  ; CHECK: cost of 1 {{.*}} %I16 = sdiv
+  %I16 = sdiv i16 undef, 7
+  ; SSE: cost of 6 {{.*}} %V8i16 = sdiv
+  ; AVX: cost of 6 {{.*}} %V8i16 = sdiv
+  %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ; SSE: cost of 12 {{.*}} %V16i16 = sdiv
+  ; AVX1: cost of 320 {{.*}} %V16i16 = sdiv
+  ; AVX2: cost of 6 {{.*}} %V16i16 = sdiv
+  ; AVX512: cost of 6 {{.*}} %V16i16 = sdiv
+  %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ; SSE: cost of 24 {{.*}} %V32i16 = sdiv
+  ; AVX1: cost of 640 {{.*}} %V32i16 = sdiv
+  ; AVX2: cost of 12 {{.*}} %V32i16 = sdiv
+  ; AVX512F: cost of 12 {{.*}} %V32i16 = sdiv
+  ; AVX512BW: cost of 640 {{.*}} %V32i16 = sdiv
+  %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+
+  ; CHECK: cost of 1 {{.*}} %I8 = sdiv
+  %I8 = sdiv i8 undef, 7
+  ; SSE: cost of 320 {{.*}} %V16i8 = sdiv
+  ; AVX: cost of 320 {{.*}} %V16i8 = sdiv
+  %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  ; SSE: cost of 640 {{.*}} %V32i8 = sdiv
+  ; AVX: cost of 640 {{.*}} %V32i8 = sdiv
+  %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  ; SSE: cost of 1280 {{.*}} %V64i8 = sdiv
+  ; AVX: cost of 1280 {{.*}} %V64i8 = sdiv
+  %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+
+  ret i32 undef
+}\r
+\r
+; CHECK-LABEL: 'udiv_uniformconst'
+define i32 @udiv_uniformconst() {
+  ; CHECK: cost of 1 {{.*}} %I64 = udiv
+  %I64 = udiv i64 undef, 7
+  ; SSE: cost of 40 {{.*}} %V2i64 = udiv
+  ; AVX: cost of 40 {{.*}} %V2i64 = udiv
+  %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+  ; SSE: cost of 80 {{.*}} %V4i64 = udiv
+  ; AVX: cost of 80 {{.*}} %V4i64 = udiv
+  %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+  ; SSE: cost of 160 {{.*}} %V8i64 = udiv
+  ; AVX: cost of 160 {{.*}} %V8i64 = udiv
+  %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+
+  ; CHECK: cost of 1 {{.*}} %I32 = udiv
+  %I32 = udiv i32 undef, 7
+  ; SSE: cost of 15 {{.*}} %V4i32 = udiv
+  ; AVX: cost of 15 {{.*}} %V4i32 = udiv
+  %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+  ; SSE: cost of 30 {{.*}} %V8i32 = udiv
+  ; AVX1: cost of 160 {{.*}} %V8i32 = udiv
+  ; AVX2: cost of 15 {{.*}} %V8i32 = udiv
+  ; AVX512: cost of 15 {{.*}} %V8i32 = udiv
+  %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+  ; SSE: cost of 60 {{.*}} %V16i32 = udiv
+  ; AVX1: cost of 320 {{.*}} %V16i32 = udiv
+  ; AVX2: cost of 30 {{.*}} %V16i32 = udiv
+  ; AVX512F: cost of 48 {{.*}} %V16i32 = udiv
+  ; AVX512BW: cost of 320 {{.*}} %V16i32 = udiv
+  %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+
+  ; CHECK: cost of 1 {{.*}} %I16 = udiv
+  %I16 = udiv i16 undef, 7
+  ; SSE: cost of 6 {{.*}} %V8i16 = udiv
+  ; AVX: cost of 6 {{.*}} %V8i16 = udiv
+  %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ; SSE: cost of 12 {{.*}} %V16i16 = udiv
+  ; AVX1: cost of 320 {{.*}} %V16i16 = udiv
+  ; AVX2: cost of 6 {{.*}} %V16i16 = udiv
+  ; AVX512: cost of 6 {{.*}} %V16i16 = udiv
+  %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ; SSE: cost of 24 {{.*}} %V32i16 = udiv
+  ; AVX1: cost of 640 {{.*}} %V32i16 = udiv
+  ; AVX2: cost of 12 {{.*}} %V32i16 = udiv
+  ; AVX512F: cost of 12 {{.*}} %V32i16 = udiv
+  ; AVX512BW: cost of 640 {{.*}} %V32i16 = udiv
+  %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+
+  ; CHECK: cost of 1 {{.*}} %I8 = udiv
+  %I8 = udiv i8 undef, 7
+  ; SSE: cost of 320 {{.*}} %V16i8 = udiv
+  ; AVX: cost of 320 {{.*}} %V16i8 = udiv
+  %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  ; SSE: cost of 640 {{.*}} %V32i8 = udiv
+  ; AVX: cost of 640 {{.*}} %V32i8 = udiv
+  %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+  ; SSE: cost of 1280 {{.*}} %V64i8 = udiv
+  ; AVX: cost of 1280 {{.*}} %V64i8 = udiv
+  %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+
+  ret i32 undef
+}\r
+\r
+; CHECK-LABEL: 'sdiv_uniformconstpow2'
+define i32 @sdiv_uniformconstpow2() {
+  ; CHECK: cost of 1 {{.*}} %I64 = sdiv
+  %I64 = sdiv i64 undef, 16
+  ; SSE: cost of 40 {{.*}} %V2i64 = sdiv
+  ; AVX: cost of 40 {{.*}} %V2i64 = sdiv
+  %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+  ; SSE: cost of 80 {{.*}} %V4i64 = sdiv
+  ; AVX: cost of 80 {{.*}} %V4i64 = sdiv
+  %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+  ; SSE: cost of 160 {{.*}} %V8i64 = sdiv
+  ; AVX: cost of 160 {{.*}} %V8i64 = sdiv
+  %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+
+  ; CHECK: cost of 1 {{.*}} %I32 = sdiv
+  %I32 = sdiv i32 undef, 16
+  ; SSE2: cost of 19 {{.*}} %V4i32 = sdiv
+  ; SSSE3: cost of 19 {{.*}} %V4i32 = sdiv
+  ; SSE42: cost of 15 {{.*}} %V4i32 = sdiv
+  ; AVX: cost of 15 {{.*}} %V4i32 = sdiv
+  %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+  ; SSE2: cost of 38 {{.*}} %V8i32 = sdiv
+  ; SSSE3: cost of 38 {{.*}} %V8i32 = sdiv
+  ; SSE42: cost of 30 {{.*}} %V8i32 = sdiv
+  ; AVX1: cost of 160 {{.*}} %V8i32 = sdiv
+  ; AVX2: cost of 15 {{.*}} %V8i32 = sdiv
+  ; AVX512: cost of 15 {{.*}} %V8i32 = sdiv
+  %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+  ; SSE2: cost of 76 {{.*}} %V16i32 = sdiv
+  ; SSSE3: cost of 76 {{.*}} %V16i32 = sdiv
+  ; SSE42: cost of 60 {{.*}} %V16i32 = sdiv
+  ; AVX1: cost of 320 {{.*}} %V16i32 = sdiv
+  ; AVX2: cost of 30 {{.*}} %V16i32 = sdiv
+  ; AVX512F: cost of 48 {{.*}} %V16i32 = sdiv
+  ; AVX512BW: cost of 320 {{.*}} %V16i32 = sdiv
+  %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+
+  ; CHECK: cost of 1 {{.*}} %I16 = sdiv
+  %I16 = sdiv i16 undef, 16
+  ; SSE: cost of 6 {{.*}} %V8i16 = sdiv
+  ; AVX: cost of 6 {{.*}} %V8i16 = sdiv
+  %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  ; SSE: cost of 12 {{.*}} %V16i16 = sdiv
+  ; AVX1: cost of 320 {{.*}} %V16i16 = sdiv
+  ; AVX2: cost of 6 {{.*}} %V16i16 = sdiv
+  ; AVX512: cost of 6 {{.*}} %V16i16 = sdiv
+  %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  ; SSE: cost of 24 {{.*}} %V32i16 = sdiv
+  ; AVX1: cost of 640 {{.*}} %V32i16 = sdiv
+  ; AVX2: cost of 12 {{.*}} %V32i16 = sdiv
+  ; AVX512F: cost of 12 {{.*}} %V32i16 = sdiv
+  ; AVX512BW: cost of 640 {{.*}} %V32i16 = sdiv
+  %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+
+  ; CHECK: cost of 1 {{.*}} %I8 = sdiv
+  %I8 = sdiv i8 undef, 16
+  ; SSE: cost of 320 {{.*}} %V16i8 = sdiv
+  ; AVX: cost of 320 {{.*}} %V16i8 = sdiv
+  %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+  ; SSE: cost of 640 {{.*}} %V32i8 = sdiv
+  ; AVX: cost of 640 {{.*}} %V32i8 = sdiv
+  %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+  ; SSE: cost of 1280 {{.*}} %V64i8 = sdiv
+  ; AVX: cost of 1280 {{.*}} %V64i8 = sdiv
+  %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+
+  ret i32 undef
+}
+\r
+; CHECK-LABEL: 'udiv_uniformconstpow2'
+define i32 @udiv_uniformconstpow2() {
+  ; CHECK: cost of 1 {{.*}} %I64 = udiv
+  %I64 = udiv i64 undef, 16
+  ; SSE: cost of 40 {{.*}} %V2i64 = udiv
+  ; AVX: cost of 40 {{.*}} %V2i64 = udiv
+  %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+  ; SSE: cost of 80 {{.*}} %V4i64 = udiv
+  ; AVX: cost of 80 {{.*}} %V4i64 = udiv
+  %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+  ; SSE: cost of 160 {{.*}} %V8i64 = udiv
+  ; AVX: cost of 160 {{.*}} %V8i64 = udiv
+  %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+
+  ; CHECK: cost of 1 {{.*}} %I32 = udiv
+  %I32 = udiv i32 undef, 16
+  ; SSE: cost of 15 {{.*}} %V4i32 = udiv
+  ; AVX: cost of 15 {{.*}} %V4i32 = udiv
+  %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+  ; SSE: cost of 30 {{.*}} %V8i32 = udiv
+  ; AVX1: cost of 160 {{.*}} %V8i32 = udiv
+  ; AVX2: cost of 15 {{.*}} %V8i32 = udiv
+  ; AVX512: cost of 15 {{.*}} %V8i32 = udiv
+  %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+  ; SSE: cost of 60 {{.*}} %V16i32 = udiv
+  ; AVX1: cost of 320 {{.*}} %V16i32 = udiv
+  ; AVX2: cost of 30 {{.*}} %V16i32 = udiv
+  ; AVX512F: cost of 48 {{.*}} %V16i32 = udiv
+  ; AVX512BW: cost of 320 {{.*}} %V16i32 = udiv
+  %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+
+  ; CHECK: cost of 1 {{.*}} %I16 = udiv
+  %I16 = udiv i16 undef, 16
+  ; SSE: cost of 6 {{.*}} %V8i16 = udiv
+  ; AVX: cost of 6 {{.*}} %V8i16 = udiv
+  %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  ; SSE: cost of 12 {{.*}} %V16i16 = udiv
+  ; AVX1: cost of 320 {{.*}} %V16i16 = udiv
+  ; AVX2: cost of 6 {{.*}} %V16i16 = udiv
+  ; AVX512: cost of 6 {{.*}} %V16i16 = udiv
+  %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+  ; SSE: cost of 24 {{.*}} %V32i16 = udiv
+  ; AVX1: cost of 640 {{.*}} %V32i16 = udiv
+  ; AVX2: cost of 12 {{.*}} %V32i16 = udiv
+  ; AVX512F: cost of 12 {{.*}} %V32i16 = udiv
+  ; AVX512BW: cost of 640 {{.*}} %V32i16 = udiv
+  %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+
+  ; CHECK: cost of 1 {{.*}} %I8 = udiv
+  %I8 = udiv i8 undef, 16
+  ; SSE: cost of 320 {{.*}} %V16i8 = udiv
+  ; AVX: cost of 320 {{.*}} %V16i8 = udiv
+  %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+  ; SSE: cost of 640 {{.*}} %V32i8 = udiv
+  ; AVX: cost of 640 {{.*}} %V32i8 = udiv
+  %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+  ; SSE: cost of 1280 {{.*}} %V64i8 = udiv
+  ; AVX: cost of 1280 {{.*}} %V64i8 = udiv
+  %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+
+  ret i32 undef
+}\r