1291101aaSYolanda Chen; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2 2291101aaSYolanda Chen; RUN: llc < %s -verify-machineinstrs -mattr=+simd128 | FileCheck %s 3291101aaSYolanda Chen 4291101aaSYolanda Chen; Test that SIMD shifts can be lowered correctly even when shift 5291101aaSYolanda Chen; values are exported from outside blocks. 6291101aaSYolanda Chen 7291101aaSYolanda Chentarget triple = "wasm32-unknown-unknown" 8291101aaSYolanda Chen 9291101aaSYolanda Chendefine void @shl_loop(ptr %a, i8 %shift, i32 %count) { 10291101aaSYolanda Chen; CHECK-LABEL: shl_loop: 11291101aaSYolanda Chen; CHECK: .functype shl_loop (i32, i32, i32) -> () 12*4b3ba64bSNikita Popov; CHECK-NEXT: .local i32 13291101aaSYolanda Chen; CHECK-NEXT: # %bb.0: # %entry 14291101aaSYolanda Chen; CHECK-NEXT: .LBB0_1: # %body 15291101aaSYolanda Chen; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 16291101aaSYolanda Chen; CHECK-NEXT: loop # label0: 17291101aaSYolanda Chen; CHECK-NEXT: local.get 0 18*4b3ba64bSNikita Popov; CHECK-NEXT: i32.const 16 19*4b3ba64bSNikita Popov; CHECK-NEXT: i32.add 20*4b3ba64bSNikita Popov; CHECK-NEXT: local.tee 3 21291101aaSYolanda Chen; CHECK-NEXT: local.get 0 22291101aaSYolanda Chen; CHECK-NEXT: v128.load 0:p2align=0 23291101aaSYolanda Chen; CHECK-NEXT: local.get 1 24291101aaSYolanda Chen; CHECK-NEXT: i8x16.shl 25*4b3ba64bSNikita Popov; CHECK-NEXT: v128.store 0 26*4b3ba64bSNikita Popov; CHECK-NEXT: local.get 3 27291101aaSYolanda Chen; CHECK-NEXT: local.set 0 28291101aaSYolanda Chen; CHECK-NEXT: local.get 2 29291101aaSYolanda Chen; CHECK-NEXT: i32.const -1 30291101aaSYolanda Chen; CHECK-NEXT: i32.add 31291101aaSYolanda Chen; CHECK-NEXT: local.tee 2 32291101aaSYolanda Chen; CHECK-NEXT: i32.eqz 33291101aaSYolanda Chen; CHECK-NEXT: br_if 0 # 0: up to label0 34291101aaSYolanda Chen; CHECK-NEXT: # %bb.2: # %exit 35291101aaSYolanda Chen; CHECK-NEXT: end_loop 36291101aaSYolanda Chen; CHECK-NEXT: # fallthrough-return 37291101aaSYolanda Chenentry: 38291101aaSYolanda Chen %t1 = insertelement <16 x i8> undef, i8 %shift, i32 0 39291101aaSYolanda Chen %vshift = shufflevector <16 x i8> %t1, <16 x i8> undef, <16 x i32> zeroinitializer 40291101aaSYolanda Chen br label %body 41291101aaSYolanda Chenbody: 42291101aaSYolanda Chen %out = phi ptr [%a, %entry], [%b, %body] 43291101aaSYolanda Chen %i = phi i32 [0, %entry], [%next, %body] 44291101aaSYolanda Chen %v = load <16 x i8>, ptr %out, align 1 45291101aaSYolanda Chen %r = shl <16 x i8> %v, %vshift 46291101aaSYolanda Chen %b = getelementptr inbounds i8, ptr %out, i32 16 47291101aaSYolanda Chen store <16 x i8> %r, ptr %b 48291101aaSYolanda Chen %next = add i32 %i, 1 49291101aaSYolanda Chen %i.cmp = icmp eq i32 %next, %count 50291101aaSYolanda Chen br i1 %i.cmp, label %body, label %exit 51291101aaSYolanda Chenexit: 52291101aaSYolanda Chen ret void 53291101aaSYolanda Chen} 54291101aaSYolanda Chen 55291101aaSYolanda Chen; Test that SIMD shifts can be lowered correctly when shift value 56291101aaSYolanda Chen; is a phi inside loop body. 57291101aaSYolanda Chen 58291101aaSYolanda Chendefine void @shl_phi_loop(ptr %a, i8 %shift, i32 %count) { 59291101aaSYolanda Chen; CHECK-LABEL: shl_phi_loop: 60291101aaSYolanda Chen; CHECK: .functype shl_phi_loop (i32, i32, i32) -> () 61*4b3ba64bSNikita Popov; CHECK-NEXT: .local i32 62291101aaSYolanda Chen; CHECK-NEXT: # %bb.0: # %entry 63291101aaSYolanda Chen; CHECK-NEXT: .LBB1_1: # %body 64291101aaSYolanda Chen; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 65291101aaSYolanda Chen; CHECK-NEXT: loop # label1: 66291101aaSYolanda Chen; CHECK-NEXT: local.get 0 67*4b3ba64bSNikita Popov; CHECK-NEXT: i32.const 16 68*4b3ba64bSNikita Popov; CHECK-NEXT: i32.add 69*4b3ba64bSNikita Popov; CHECK-NEXT: local.tee 3 70291101aaSYolanda Chen; CHECK-NEXT: local.get 0 71291101aaSYolanda Chen; CHECK-NEXT: v128.load 0:p2align=0 72291101aaSYolanda Chen; CHECK-NEXT: local.get 1 73291101aaSYolanda Chen; CHECK-NEXT: i8x16.shl 74*4b3ba64bSNikita Popov; CHECK-NEXT: v128.store 0 75291101aaSYolanda Chen; CHECK-NEXT: local.get 1 76291101aaSYolanda Chen; CHECK-NEXT: i32.const 1 77291101aaSYolanda Chen; CHECK-NEXT: i32.and 78291101aaSYolanda Chen; CHECK-NEXT: local.set 1 79*4b3ba64bSNikita Popov; CHECK-NEXT: local.get 3 80291101aaSYolanda Chen; CHECK-NEXT: local.set 0 81291101aaSYolanda Chen; CHECK-NEXT: local.get 2 82291101aaSYolanda Chen; CHECK-NEXT: i32.const -1 83291101aaSYolanda Chen; CHECK-NEXT: i32.add 84291101aaSYolanda Chen; CHECK-NEXT: local.tee 2 85291101aaSYolanda Chen; CHECK-NEXT: i32.eqz 86291101aaSYolanda Chen; CHECK-NEXT: br_if 0 # 0: up to label1 87291101aaSYolanda Chen; CHECK-NEXT: # %bb.2: # %exit 88291101aaSYolanda Chen; CHECK-NEXT: end_loop 89291101aaSYolanda Chen; CHECK-NEXT: # fallthrough-return 90291101aaSYolanda Chenentry: 91291101aaSYolanda Chen br label %body 92291101aaSYolanda Chenbody: 93291101aaSYolanda Chen %out = phi ptr [%a, %entry], [%b, %body] 94291101aaSYolanda Chen %i = phi i32 [0, %entry], [%next, %body] 95291101aaSYolanda Chen %t1 = phi i8 [%shift, %entry], [%sand, %body] 96291101aaSYolanda Chen %t2 = insertelement <16 x i8> undef, i8 %t1, i32 0 97291101aaSYolanda Chen %vshift = shufflevector <16 x i8> %t2, <16 x i8> undef, <16 x i32> zeroinitializer 98291101aaSYolanda Chen %v = load <16 x i8>, ptr %out, align 1 99291101aaSYolanda Chen %r = shl <16 x i8> %v, %vshift 100291101aaSYolanda Chen %b = getelementptr inbounds i8, ptr %out, i32 16 101291101aaSYolanda Chen store <16 x i8> %r, ptr %b 102291101aaSYolanda Chen %sand = and i8 %t1, 1 103291101aaSYolanda Chen %next = add i32 %i, 1 104291101aaSYolanda Chen %i.cmp = icmp eq i32 %next, %count 105291101aaSYolanda Chen br i1 %i.cmp, label %body, label %exit 106291101aaSYolanda Chenexit: 107291101aaSYolanda Chen ret void 108291101aaSYolanda Chen} 109