1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mattr=+simd128 | FileCheck %s 3 4;; Test that SIMD widening operations can be successfully selected 5 6target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128" 7target triple = "wasm32-unknown-unknown" 8 9define <8 x i16> @widen_low_i8x16_s(<16 x i8> %v) { 10; CHECK-LABEL: widen_low_i8x16_s: 11; CHECK: .functype widen_low_i8x16_s (v128) -> (v128) 12; CHECK-NEXT: # %bb.0: 13; CHECK-NEXT: local.get 0 14; CHECK-NEXT: i16x8.widen_low_i8x16_s 15; CHECK-NEXT: # fallthrough-return 16 %low = shufflevector <16 x i8> %v, <16 x i8> undef, 17 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 18 %widened = sext <8 x i8> %low to <8 x i16> 19 ret <8 x i16> %widened 20} 21 22define <8 x i16> @widen_low_i8x16_u(<16 x i8> %v) { 23; CHECK-LABEL: widen_low_i8x16_u: 24; CHECK: .functype widen_low_i8x16_u (v128) -> (v128) 25; CHECK-NEXT: # %bb.0: 26; CHECK-NEXT: local.get 0 27; CHECK-NEXT: i16x8.widen_low_i8x16_u 28; CHECK-NEXT: # fallthrough-return 29 %low = shufflevector <16 x i8> %v, <16 x i8> undef, 30 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 31 %widened = zext <8 x i8> %low to <8 x i16> 32 ret <8 x i16> %widened 33} 34 35define <8 x i16> @widen_high_i8x16_s(<16 x i8> %v) { 36; CHECK-LABEL: widen_high_i8x16_s: 37; CHECK: .functype widen_high_i8x16_s (v128) -> (v128) 38; CHECK-NEXT: # %bb.0: 39; CHECK-NEXT: local.get 0 40; CHECK-NEXT: i16x8.widen_high_i8x16_s 41; CHECK-NEXT: # fallthrough-return 42 %low = shufflevector <16 x i8> %v, <16 x i8> undef, 43 <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> 44 %widened = sext <8 x i8> %low to <8 x i16> 45 ret <8 x i16> %widened 46} 47 48define <8 x i16> @widen_high_i8x16_u(<16 x i8> %v) { 49; CHECK-LABEL: widen_high_i8x16_u: 50; CHECK: .functype widen_high_i8x16_u (v128) -> (v128) 51; CHECK-NEXT: # %bb.0: 52; CHECK-NEXT: local.get 0 53; CHECK-NEXT: i16x8.widen_high_i8x16_u 54; CHECK-NEXT: # fallthrough-return 55 %low = shufflevector <16 x i8> %v, <16 x i8> undef, 56 <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> 57 %widened = zext <8 x i8> %low to <8 x i16> 58 ret <8 x i16> %widened 59} 60 61define <4 x i32> @widen_low_i16x8_s(<8 x i16> %v) { 62; CHECK-LABEL: widen_low_i16x8_s: 63; CHECK: .functype widen_low_i16x8_s (v128) -> (v128) 64; CHECK-NEXT: # %bb.0: 65; CHECK-NEXT: local.get 0 66; CHECK-NEXT: i32x4.widen_low_i16x8_s 67; CHECK-NEXT: # fallthrough-return 68 %low = shufflevector <8 x i16> %v, <8 x i16> undef, 69 <4 x i32> <i32 0, i32 1, i32 2, i32 3> 70 %widened = sext <4 x i16> %low to <4 x i32> 71 ret <4 x i32> %widened 72} 73 74define <4 x i32> @widen_low_i16x8_u(<8 x i16> %v) { 75; CHECK-LABEL: widen_low_i16x8_u: 76; CHECK: .functype widen_low_i16x8_u (v128) -> (v128) 77; CHECK-NEXT: # %bb.0: 78; CHECK-NEXT: local.get 0 79; CHECK-NEXT: i32x4.widen_low_i16x8_u 80; CHECK-NEXT: # fallthrough-return 81 %low = shufflevector <8 x i16> %v, <8 x i16> undef, 82 <4 x i32> <i32 0, i32 1, i32 2, i32 3> 83 %widened = zext <4 x i16> %low to <4 x i32> 84 ret <4 x i32> %widened 85} 86 87define <4 x i32> @widen_high_i16x8_s(<8 x i16> %v) { 88; CHECK-LABEL: widen_high_i16x8_s: 89; CHECK: .functype widen_high_i16x8_s (v128) -> (v128) 90; CHECK-NEXT: # %bb.0: 91; CHECK-NEXT: local.get 0 92; CHECK-NEXT: i32x4.widen_high_i16x8_s 93; CHECK-NEXT: # fallthrough-return 94 %low = shufflevector <8 x i16> %v, <8 x i16> undef, 95 <4 x i32> <i32 4, i32 5, i32 6, i32 7> 96 %widened = sext <4 x i16> %low to <4 x i32> 97 ret <4 x i32> %widened 98} 99 100define <4 x i32> @widen_high_i16x8_u(<8 x i16> %v) { 101; CHECK-LABEL: widen_high_i16x8_u: 102; CHECK: .functype widen_high_i16x8_u (v128) -> (v128) 103; CHECK-NEXT: # %bb.0: 104; CHECK-NEXT: local.get 0 105; CHECK-NEXT: i32x4.widen_high_i16x8_u 106; CHECK-NEXT: # fallthrough-return 107 %low = shufflevector <8 x i16> %v, <8 x i16> undef, 108 <4 x i32> <i32 4, i32 5, i32 6, i32 7> 109 %widened = zext <4 x i16> %low to <4 x i32> 110 ret <4 x i32> %widened 111} 112 113;; Also test that similar patterns with offsets not corresponding to 114;; the low or high half are correctly expanded. 115 116define <8 x i16> @widen_lowish_i8x16_s(<16 x i8> %v) { 117; CHECK-LABEL: widen_lowish_i8x16_s: 118; CHECK: .functype widen_lowish_i8x16_s (v128) -> (v128) 119; CHECK-NEXT: # %bb.0: 120; CHECK-NEXT: local.get 0 121; CHECK-NEXT: i8x16.extract_lane_u 1 122; CHECK-NEXT: i16x8.splat 123; CHECK-NEXT: local.get 0 124; CHECK-NEXT: i8x16.extract_lane_u 2 125; CHECK-NEXT: i16x8.replace_lane 1 126; CHECK-NEXT: local.get 0 127; CHECK-NEXT: i8x16.extract_lane_u 3 128; CHECK-NEXT: i16x8.replace_lane 2 129; CHECK-NEXT: local.get 0 130; CHECK-NEXT: i8x16.extract_lane_u 4 131; CHECK-NEXT: i16x8.replace_lane 3 132; CHECK-NEXT: local.get 0 133; CHECK-NEXT: i8x16.extract_lane_u 5 134; CHECK-NEXT: i16x8.replace_lane 4 135; CHECK-NEXT: local.get 0 136; CHECK-NEXT: i8x16.extract_lane_u 6 137; CHECK-NEXT: i16x8.replace_lane 5 138; CHECK-NEXT: local.get 0 139; CHECK-NEXT: i8x16.extract_lane_u 7 140; CHECK-NEXT: i16x8.replace_lane 6 141; CHECK-NEXT: local.get 0 142; CHECK-NEXT: i8x16.extract_lane_u 8 143; CHECK-NEXT: i16x8.replace_lane 7 144; CHECK-NEXT: i32.const 8 145; CHECK-NEXT: i16x8.shl 146; CHECK-NEXT: i32.const 8 147; CHECK-NEXT: i16x8.shr_s 148; CHECK-NEXT: # fallthrough-return 149 %lowish = shufflevector <16 x i8> %v, <16 x i8> undef, 150 <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8> 151 %widened = sext <8 x i8> %lowish to <8 x i16> 152 ret <8 x i16> %widened 153} 154 155define <4 x i32> @widen_lowish_i16x8_s(<8 x i16> %v) { 156; CHECK-LABEL: widen_lowish_i16x8_s: 157; CHECK: .functype widen_lowish_i16x8_s (v128) -> (v128) 158; CHECK-NEXT: # %bb.0: 159; CHECK-NEXT: local.get 0 160; CHECK-NEXT: i16x8.extract_lane_u 1 161; CHECK-NEXT: i32x4.splat 162; CHECK-NEXT: local.get 0 163; CHECK-NEXT: i16x8.extract_lane_u 2 164; CHECK-NEXT: i32x4.replace_lane 1 165; CHECK-NEXT: local.get 0 166; CHECK-NEXT: i16x8.extract_lane_u 3 167; CHECK-NEXT: i32x4.replace_lane 2 168; CHECK-NEXT: local.get 0 169; CHECK-NEXT: i16x8.extract_lane_u 4 170; CHECK-NEXT: i32x4.replace_lane 3 171; CHECK-NEXT: i32.const 16 172; CHECK-NEXT: i32x4.shl 173; CHECK-NEXT: i32.const 16 174; CHECK-NEXT: i32x4.shr_s 175; CHECK-NEXT: # fallthrough-return 176 %lowish = shufflevector <8 x i16> %v, <8 x i16> undef, 177 <4 x i32> <i32 1, i32 2, i32 3, i32 4> 178 %widened = sext <4 x i16> %lowish to <4 x i32> 179 ret <4 x i32> %widened 180} 181