| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s --check-prefix=CHECK --check-prefix=KNL |
| ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX |
| |
| define i16 @mask16(i16 %x) { |
| ; CHECK-LABEL: mask16: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: kmovw %edi, %k0 |
| ; CHECK-NEXT: knotw %k0, %k0 |
| ; CHECK-NEXT: kmovw %k0, %eax |
| ; CHECK-NEXT: retq |
| %m0 = bitcast i16 %x to <16 x i1> |
| %m1 = xor <16 x i1> %m0, <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1> |
| %ret = bitcast <16 x i1> %m1 to i16 |
| ret i16 %ret |
| } |
| |
| define i8 @mask8(i8 %x) { |
| ; KNL-LABEL: mask8: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: movzbl %dil, %eax |
| ; KNL-NEXT: kmovw %eax, %k0 |
| ; KNL-NEXT: knotw %k0, %k0 |
| ; KNL-NEXT: kmovw %k0, %eax |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: mask8: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovb %edi, %k0 |
| ; SKX-NEXT: knotb %k0, %k0 |
| ; SKX-NEXT: kmovb %k0, %eax |
| ; SKX-NEXT: retq |
| %m0 = bitcast i8 %x to <8 x i1> |
| %m1 = xor <8 x i1> %m0, <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1> |
| %ret = bitcast <8 x i1> %m1 to i8 |
| ret i8 %ret |
| } |
| |
| define void @mask16_mem(i16* %ptr) { |
| ; CHECK-LABEL: mask16_mem: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: kmovw (%rdi), %k0 |
| ; CHECK-NEXT: knotw %k0, %k0 |
| ; CHECK-NEXT: kmovw %k0, (%rdi) |
| ; CHECK-NEXT: retq |
| %x = load i16, i16* %ptr, align 4 |
| %m0 = bitcast i16 %x to <16 x i1> |
| %m1 = xor <16 x i1> %m0, <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1> |
| %ret = bitcast <16 x i1> %m1 to i16 |
| store i16 %ret, i16* %ptr, align 4 |
| ret void |
| } |
| |
| define void @mask8_mem(i8* %ptr) { |
| ; KNL-LABEL: mask8_mem: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: kmovw (%rdi), %k0 |
| ; KNL-NEXT: knotw %k0, %k0 |
| ; KNL-NEXT: kmovw %k0, (%rdi) |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: mask8_mem: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovb (%rdi), %k0 |
| ; SKX-NEXT: knotb %k0, %k0 |
| ; SKX-NEXT: kmovb %k0, (%rdi) |
| ; SKX-NEXT: retq |
| %x = load i8, i8* %ptr, align 4 |
| %m0 = bitcast i8 %x to <8 x i1> |
| %m1 = xor <8 x i1> %m0, <i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1, i1 -1> |
| %ret = bitcast <8 x i1> %m1 to i8 |
| store i8 %ret, i8* %ptr, align 4 |
| ret void |
| } |
| |
| define i16 @mand16(i16 %x, i16 %y) { |
| ; CHECK-LABEL: mand16: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: kmovw %edi, %k0 |
| ; CHECK-NEXT: kmovw %esi, %k1 |
| ; CHECK-NEXT: kandw %k1, %k0, %k2 |
| ; CHECK-NEXT: kxorw %k1, %k0, %k0 |
| ; CHECK-NEXT: korw %k0, %k2, %k0 |
| ; CHECK-NEXT: kmovw %k0, %eax |
| ; CHECK-NEXT: retq |
| %ma = bitcast i16 %x to <16 x i1> |
| %mb = bitcast i16 %y to <16 x i1> |
| %mc = and <16 x i1> %ma, %mb |
| %md = xor <16 x i1> %ma, %mb |
| %me = or <16 x i1> %mc, %md |
| %ret = bitcast <16 x i1> %me to i16 |
| ret i16 %ret |
| } |
| |
| define i8 @shuf_test1(i16 %v) nounwind { |
| ; KNL-LABEL: shuf_test1: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: kmovw %edi, %k0 |
| ; KNL-NEXT: kshiftrw $8, %k0, %k0 |
| ; KNL-NEXT: kmovw %k0, %eax |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: shuf_test1: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovw %edi, %k0 |
| ; SKX-NEXT: kshiftrw $8, %k0, %k0 |
| ; SKX-NEXT: kmovb %k0, %eax |
| ; SKX-NEXT: retq |
| %v1 = bitcast i16 %v to <16 x i1> |
| %mask = shufflevector <16 x i1> %v1, <16 x i1> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> |
| %mask1 = bitcast <8 x i1> %mask to i8 |
| ret i8 %mask1 |
| } |
| |
| define i32 @zext_test1(<16 x i32> %a, <16 x i32> %b) { |
| ; CHECK-LABEL: zext_test1: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: vpcmpnleud %zmm1, %zmm0, %k0 |
| ; CHECK-NEXT: kshiftlw $10, %k0, %k0 |
| ; CHECK-NEXT: kshiftrw $15, %k0, %k0 |
| ; CHECK-NEXT: kmovw %k0, %eax |
| ; CHECK-NEXT: andl $1, %eax |
| ; CHECK-NEXT: retq |
| %cmp_res = icmp ugt <16 x i32> %a, %b |
| %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5 |
| %res = zext i1 %cmp_res.i1 to i32 |
| ret i32 %res |
| }define i16 @zext_test2(<16 x i32> %a, <16 x i32> %b) { |
| %cmp_res = icmp ugt <16 x i32> %a, %b |
| %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5 |
| %res = zext i1 %cmp_res.i1 to i16 |
| ret i16 %res |
| }define i8 @zext_test3(<16 x i32> %a, <16 x i32> %b) { |
| %cmp_res = icmp ugt <16 x i32> %a, %b |
| %cmp_res.i1 = extractelement <16 x i1> %cmp_res, i32 5 |
| %res = zext i1 %cmp_res.i1 to i8 |
| ret i8 %res |
| } |
| |
| define i8 @conv1(<8 x i1>* %R) { |
| ; KNL-LABEL: conv1: |
| ; KNL: ## BB#0: ## %entry |
| ; KNL-NEXT: kxnorw %k0, %k0, %k0 |
| ; KNL-NEXT: kmovw %k0, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: movb $-2, -{{[0-9]+}}(%rsp) |
| ; KNL-NEXT: movb $-2, %al |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: conv1: |
| ; SKX: ## BB#0: ## %entry |
| ; SKX-NEXT: kxnorw %k0, %k0, %k0 |
| ; SKX-NEXT: kmovb %k0, (%rdi) |
| ; SKX-NEXT: movb $-2, -{{[0-9]+}}(%rsp) |
| ; SKX-NEXT: movb $-2, %al |
| ; SKX-NEXT: retq |
| entry: |
| store <8 x i1> <i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1>, <8 x i1>* %R |
| |
| %maskPtr = alloca <8 x i1> |
| store <8 x i1> <i1 0, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1>, <8 x i1>* %maskPtr |
| %mask = load <8 x i1>, <8 x i1>* %maskPtr |
| %mask_convert = bitcast <8 x i1> %mask to i8 |
| ret i8 %mask_convert |
| } |
| |
| define <4 x i32> @test4(<4 x i64> %x, <4 x i64> %y, <4 x i64> %x1, <4 x i64> %y1) { |
| ; KNL-LABEL: test4: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpcmpgtq %ymm1, %ymm0, %ymm0 |
| ; KNL-NEXT: vpmovqd %zmm0, %ymm0 |
| ; KNL-NEXT: vpslld $31, %xmm0, %xmm0 |
| ; KNL-NEXT: vpsrad $31, %xmm0, %xmm0 |
| ; KNL-NEXT: vpcmpgtq %ymm3, %ymm2, %ymm1 |
| ; KNL-NEXT: vpmovqd %zmm1, %ymm1 |
| ; KNL-NEXT: vpslld $31, %xmm1, %xmm1 |
| ; KNL-NEXT: vpsrad $31, %xmm1, %xmm1 |
| ; KNL-NEXT: vpcmpgtd %xmm1, %xmm0, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test4: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpcmpgtq %ymm3, %ymm2, %k0 |
| ; SKX-NEXT: knotw %k0, %k1 |
| ; SKX-NEXT: vpcmpgtq %ymm1, %ymm0, %k0 {%k1} |
| ; SKX-NEXT: vpmovm2d %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %x_gt_y = icmp sgt <4 x i64> %x, %y |
| %x1_gt_y1 = icmp sgt <4 x i64> %x1, %y1 |
| %res = icmp sgt <4 x i1>%x_gt_y, %x1_gt_y1 |
| %resse = sext <4 x i1>%res to <4 x i32> |
| ret <4 x i32> %resse |
| } |
| |
| define <2 x i64> @test5(<2 x i64> %x, <2 x i64> %y, <2 x i64> %x1, <2 x i64> %y1) { |
| ; KNL-LABEL: test5: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0 |
| ; KNL-NEXT: vpcmpgtq %xmm3, %xmm2, %xmm1 |
| ; KNL-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test5: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpcmpgtq %xmm0, %xmm1, %k0 |
| ; SKX-NEXT: knotw %k0, %k1 |
| ; SKX-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} |
| ; SKX-NEXT: vpmovm2q %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %x_gt_y = icmp slt <2 x i64> %x, %y |
| %x1_gt_y1 = icmp sgt <2 x i64> %x1, %y1 |
| %res = icmp slt <2 x i1>%x_gt_y, %x1_gt_y1 |
| %resse = sext <2 x i1>%res to <2 x i64> |
| ret <2 x i64> %resse |
| }define void @test6(<16 x i1> %mask) { |
| allocas: |
| %a= and <16 x i1> %mask, <i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false> |
| %b = bitcast <16 x i1> %a to i16 |
| %c = icmp eq i16 %b, 0 |
| br i1 %c, label %true, label %false |
| |
| true: |
| ret void |
| |
| false: |
| ret void |
| } |
| define void @test7(<8 x i1> %mask) { |
| ; KNL-LABEL: test7: |
| ; KNL: ## BB#0: ## %allocas |
| ; KNL-NEXT: vpmovsxwq %xmm0, %zmm0 |
| ; KNL-NEXT: vpsllq $63, %zmm0, %zmm0 |
| ; KNL-NEXT: vptestmq %zmm0, %zmm0, %k0 |
| ; KNL-NEXT: movb $85, %al |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: kmovw %eax, %k1 |
| ; KNL-NEXT: korw %k1, %k0, %k0 |
| ; KNL-NEXT: kmovw %k0, %eax |
| ; KNL-NEXT: testb %al, %al |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test7: |
| ; SKX: ## BB#0: ## %allocas |
| ; SKX-NEXT: vpsllw $15, %xmm0, %xmm0 |
| ; SKX-NEXT: vpmovw2m %xmm0, %k0 |
| ; SKX-NEXT: movb $85, %al |
| ; SKX-NEXT: kmovb %eax, %k1 |
| ; SKX-NEXT: korb %k1, %k0, %k0 |
| ; SKX-NEXT: kmovb %k0, %eax |
| ; SKX-NEXT: testb %al, %al |
| ; SKX-NEXT: retq |
| allocas: |
| %a= or <8 x i1> %mask, <i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false> |
| %b = bitcast <8 x i1> %a to i8 |
| %c = icmp eq i8 %b, 0 |
| br i1 %c, label %true, label %false |
| |
| true: |
| ret void |
| |
| false: |
| ret void |
| } |
| define <16 x i8> @test8(<16 x i32>%a, <16 x i32>%b, i32 %a1, i32 %b1) { |
| ; KNL-LABEL: test8: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpxord %zmm2, %zmm2, %zmm2 |
| ; KNL-NEXT: cmpl %esi, %edi |
| ; KNL-NEXT: jg LBB14_1 |
| ; KNL-NEXT: ## BB#2: |
| ; KNL-NEXT: vpcmpltud %zmm2, %zmm1, %k1 |
| ; KNL-NEXT: jmp LBB14_3 |
| ; KNL-NEXT: LBB14_1: |
| ; KNL-NEXT: vpcmpgtd %zmm2, %zmm0, %k1 |
| ; KNL-NEXT: LBB14_3: |
| ; KNL-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} |
| ; KNL-NEXT: vpmovdb %zmm0, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test8: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpxord %zmm2, %zmm2, %zmm2 |
| ; SKX-NEXT: cmpl %esi, %edi |
| ; SKX-NEXT: jg LBB14_1 |
| ; SKX-NEXT: ## BB#2: |
| ; SKX-NEXT: vpcmpltud %zmm2, %zmm1, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %xmm0 |
| ; SKX-NEXT: retq |
| ; SKX-NEXT: LBB14_1: |
| ; SKX-NEXT: vpcmpgtd %zmm2, %zmm0, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %cond = icmp sgt i32 %a1, %b1 |
| %cmp1 = icmp sgt <16 x i32> %a, zeroinitializer |
| %cmp2 = icmp ult <16 x i32> %b, zeroinitializer |
| %mix = select i1 %cond, <16 x i1> %cmp1, <16 x i1> %cmp2 |
| %res = sext <16 x i1> %mix to <16 x i8> |
| ret <16 x i8> %res |
| } |
| define <16 x i1> @test9(<16 x i1>%a, <16 x i1>%b, i32 %a1, i32 %b1) { |
| ; KNL-LABEL: test9: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: cmpl %esi, %edi |
| ; KNL-NEXT: jg LBB15_1 |
| ; KNL-NEXT: ## BB#2: |
| ; KNL-NEXT: vpmovsxbd %xmm1, %zmm0 |
| ; KNL-NEXT: jmp LBB15_3 |
| ; KNL-NEXT: LBB15_1: |
| ; KNL-NEXT: vpmovsxbd %xmm0, %zmm0 |
| ; KNL-NEXT: LBB15_3: |
| ; KNL-NEXT: vpslld $31, %zmm0, %zmm0 |
| ; KNL-NEXT: vptestmd %zmm0, %zmm0, %k1 |
| ; KNL-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} |
| ; KNL-NEXT: vpmovdb %zmm0, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test9: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: cmpl %esi, %edi |
| ; SKX-NEXT: jg LBB15_1 |
| ; SKX-NEXT: ## BB#2: |
| ; SKX-NEXT: vpsllw $7, %xmm1, %xmm0 |
| ; SKX-NEXT: jmp LBB15_3 |
| ; SKX-NEXT: LBB15_1: |
| ; SKX-NEXT: vpsllw $7, %xmm0, %xmm0 |
| ; SKX-NEXT: LBB15_3: |
| ; SKX-NEXT: vpmovb2m %xmm0, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %mask = icmp sgt i32 %a1, %b1 |
| %c = select i1 %mask, <16 x i1>%a, <16 x i1>%b |
| ret <16 x i1>%c |
| }define <8 x i1> @test10(<8 x i1>%a, <8 x i1>%b, i32 %a1, i32 %b1) { |
| %mask = icmp sgt i32 %a1, %b1 |
| %c = select i1 %mask, <8 x i1>%a, <8 x i1>%b |
| ret <8 x i1>%c |
| } |
| |
| define <4 x i1> @test11(<4 x i1>%a, <4 x i1>%b, i32 %a1, i32 %b1) { |
| ; KNL-LABEL: test11: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: cmpl %esi, %edi |
| ; KNL-NEXT: jg LBB17_2 |
| ; KNL-NEXT: ## BB#1: |
| ; KNL-NEXT: vmovaps %zmm1, %zmm0 |
| ; KNL-NEXT: LBB17_2: |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test11: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: cmpl %esi, %edi |
| ; SKX-NEXT: jg LBB17_1 |
| ; SKX-NEXT: ## BB#2: |
| ; SKX-NEXT: vpslld $31, %xmm1, %xmm0 |
| ; SKX-NEXT: jmp LBB17_3 |
| ; SKX-NEXT: LBB17_1: |
| ; SKX-NEXT: vpslld $31, %xmm0, %xmm0 |
| ; SKX-NEXT: LBB17_3: |
| ; SKX-NEXT: vpmovd2m %xmm0, %k0 |
| ; SKX-NEXT: vpmovm2d %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %mask = icmp sgt i32 %a1, %b1 |
| %c = select i1 %mask, <4 x i1>%a, <4 x i1>%b |
| ret <4 x i1>%c |
| } |
| |
| define i32 @test12(i32 %x, i32 %y) { |
| ; CHECK-LABEL: test12: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: movl %edi, %eax |
| ; CHECK-NEXT: retq |
| %a = bitcast i16 21845 to <16 x i1> |
| %b = extractelement <16 x i1> %a, i32 0 |
| %c = select i1 %b, i32 %x, i32 %y |
| ret i32 %c |
| } |
| |
| define i32 @test13(i32 %x, i32 %y) { |
| ; CHECK-LABEL: test13: |
| ; CHECK: ## BB#0: |
| ; CHECK-NEXT: movl %esi, %eax |
| ; CHECK-NEXT: retq |
| %a = bitcast i16 21845 to <16 x i1> |
| %b = extractelement <16 x i1> %a, i32 3 |
| %c = select i1 %b, i32 %x, i32 %y |
| ret i32 %c |
| }define <4 x i1> @test14() { |
| %a = bitcast i16 21845 to <16 x i1> |
| %b = extractelement <16 x i1> %a, i32 2 |
| %c = insertelement <4 x i1> <i1 true, i1 false, i1 false, i1 true>, i1 %b, i32 1 |
| ret <4 x i1> %c |
| } |
| |
| define <16 x i1> @test15(i32 %x, i32 %y) { |
| ; KNL-LABEL: test15: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: cmpl %esi, %edi |
| ; KNL-NEXT: movw $21845, %ax ## imm = 0x5555 |
| ; KNL-NEXT: movw $1, %cx |
| ; KNL-NEXT: cmovgw %ax, %cx |
| ; KNL-NEXT: kmovw %ecx, %k1 |
| ; KNL-NEXT: vpbroadcastd {{.*}}(%rip), %zmm0 {%k1} {z} |
| ; KNL-NEXT: vpmovdb %zmm0, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test15: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: cmpl %esi, %edi |
| ; SKX-NEXT: movw $21845, %ax ## imm = 0x5555 |
| ; SKX-NEXT: movw $1, %cx |
| ; SKX-NEXT: cmovgw %ax, %cx |
| ; SKX-NEXT: kmovw %ecx, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %a = bitcast i16 21845 to <16 x i1> |
| %b = bitcast i16 1 to <16 x i1> |
| %mask = icmp sgt i32 %x, %y |
| %c = select i1 %mask, <16 x i1> %a, <16 x i1> %b |
| ret <16 x i1> %c |
| } |
| |
| define <64 x i8> @test16(i64 %x) { |
| ; KNL-LABEL: test16: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: pushq %rbp |
| ; KNL-NEXT: Ltmp0: |
| ; KNL-NEXT: .cfi_def_cfa_offset 16 |
| ; KNL-NEXT: Ltmp1: |
| ; KNL-NEXT: .cfi_offset %rbp, -16 |
| ; KNL-NEXT: movq %rsp, %rbp |
| ; KNL-NEXT: Ltmp2: |
| ; KNL-NEXT: .cfi_def_cfa_register %rbp |
| ; KNL-NEXT: pushq %r15 |
| ; KNL-NEXT: pushq %r14 |
| ; KNL-NEXT: pushq %r13 |
| ; KNL-NEXT: pushq %r12 |
| ; KNL-NEXT: pushq %rbx |
| ; KNL-NEXT: andq $-32, %rsp |
| ; KNL-NEXT: subq $128, %rsp |
| ; KNL-NEXT: Ltmp3: |
| ; KNL-NEXT: .cfi_offset %rbx, -56 |
| ; KNL-NEXT: Ltmp4: |
| ; KNL-NEXT: .cfi_offset %r12, -48 |
| ; KNL-NEXT: Ltmp5: |
| ; KNL-NEXT: .cfi_offset %r13, -40 |
| ; KNL-NEXT: Ltmp6: |
| ; KNL-NEXT: .cfi_offset %r14, -32 |
| ; KNL-NEXT: Ltmp7: |
| ; KNL-NEXT: .cfi_offset %r15, -24 |
| ; KNL-NEXT: movq %rdi, %rax |
| ; KNL-NEXT: shrq $32, %rax |
| ; KNL-NEXT: movl %eax, {{[0-9]+}}(%rsp) |
| ; KNL-NEXT: movl $271, %eax ## imm = 0x10F |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: movl %edi, %ecx |
| ; KNL-NEXT: andl $1, %ecx |
| ; KNL-NEXT: vmovd %ecx, %xmm0 |
| ; KNL-NEXT: movl $257, %ecx ## imm = 0x101 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $1, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $258, %ecx ## imm = 0x102 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $2, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $259, %ecx ## imm = 0x103 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $3, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $260, %ecx ## imm = 0x104 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $4, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $261, %ecx ## imm = 0x105 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $5, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $262, %ecx ## imm = 0x106 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $6, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $263, %ecx ## imm = 0x107 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $7, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $264, %ecx ## imm = 0x108 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $8, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $265, %ecx ## imm = 0x109 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $9, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $266, %ecx ## imm = 0x10A |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $10, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $267, %ecx ## imm = 0x10B |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $11, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $268, %ecx ## imm = 0x10C |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $12, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $269, %ecx ## imm = 0x10D |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $13, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $270, %ecx ## imm = 0x10E |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: vpinsrb $14, %ecx, %xmm0, %xmm0 |
| ; KNL-NEXT: vpinsrb $15, %eax, %xmm0, %xmm1 |
| ; KNL-NEXT: movl $1, %eax |
| ; KNL-NEXT: vpinsrb $5, %eax, %xmm1, %xmm0 |
| ; KNL-NEXT: movl {{[0-9]+}}(%rsp), %r15d |
| ; KNL-NEXT: movq %r15, %rdx |
| ; KNL-NEXT: shrq $17, %rdx |
| ; KNL-NEXT: andb $1, %dl |
| ; KNL-NEXT: je LBB22_2 |
| ; KNL-NEXT: ## BB#1: |
| ; KNL-NEXT: movb $-1, %dl |
| ; KNL-NEXT: LBB22_2: |
| ; KNL-NEXT: movq %r15, %r11 |
| ; KNL-NEXT: shrq $16, %r11 |
| ; KNL-NEXT: andb $1, %r11b |
| ; KNL-NEXT: je LBB22_4 |
| ; KNL-NEXT: ## BB#3: |
| ; KNL-NEXT: movb $-1, %r11b |
| ; KNL-NEXT: LBB22_4: |
| ; KNL-NEXT: movq %r15, %r10 |
| ; KNL-NEXT: shrq $18, %r10 |
| ; KNL-NEXT: andb $1, %r10b |
| ; KNL-NEXT: je LBB22_6 |
| ; KNL-NEXT: ## BB#5: |
| ; KNL-NEXT: movb $-1, %r10b |
| ; KNL-NEXT: LBB22_6: |
| ; KNL-NEXT: movq %r15, %r9 |
| ; KNL-NEXT: shrq $19, %r9 |
| ; KNL-NEXT: andb $1, %r9b |
| ; KNL-NEXT: je LBB22_8 |
| ; KNL-NEXT: ## BB#7: |
| ; KNL-NEXT: movb $-1, %r9b |
| ; KNL-NEXT: LBB22_8: |
| ; KNL-NEXT: movq %r15, %rbx |
| ; KNL-NEXT: shrq $20, %rbx |
| ; KNL-NEXT: andb $1, %bl |
| ; KNL-NEXT: je LBB22_10 |
| ; KNL-NEXT: ## BB#9: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB22_10: |
| ; KNL-NEXT: movq %r15, %r12 |
| ; KNL-NEXT: shrq $21, %r12 |
| ; KNL-NEXT: andb $1, %r12b |
| ; KNL-NEXT: je LBB22_12 |
| ; KNL-NEXT: ## BB#11: |
| ; KNL-NEXT: movb $-1, %r12b |
| ; KNL-NEXT: LBB22_12: |
| ; KNL-NEXT: movq %r15, %r14 |
| ; KNL-NEXT: shrq $22, %r14 |
| ; KNL-NEXT: andb $1, %r14b |
| ; KNL-NEXT: je LBB22_14 |
| ; KNL-NEXT: ## BB#13: |
| ; KNL-NEXT: movb $-1, %r14b |
| ; KNL-NEXT: LBB22_14: |
| ; KNL-NEXT: movq %r15, %r8 |
| ; KNL-NEXT: shrq $23, %r8 |
| ; KNL-NEXT: andb $1, %r8b |
| ; KNL-NEXT: je LBB22_16 |
| ; KNL-NEXT: ## BB#15: |
| ; KNL-NEXT: movb $-1, %r8b |
| ; KNL-NEXT: LBB22_16: |
| ; KNL-NEXT: movq %r15, %r13 |
| ; KNL-NEXT: shrq $24, %r13 |
| ; KNL-NEXT: andb $1, %r13b |
| ; KNL-NEXT: je LBB22_18 |
| ; KNL-NEXT: ## BB#17: |
| ; KNL-NEXT: movb $-1, %r13b |
| ; KNL-NEXT: LBB22_18: |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $25, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_20 |
| ; KNL-NEXT: ## BB#19: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_20: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $26, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_22 |
| ; KNL-NEXT: ## BB#21: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_22: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movl $272, %esi ## imm = 0x110 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $27, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_24 |
| ; KNL-NEXT: ## BB#23: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_24: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movl $273, %eax ## imm = 0x111 |
| ; KNL-NEXT: bextrl %esi, %edi, %esi |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $28, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB22_26 |
| ; KNL-NEXT: ## BB#25: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB22_26: |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vmovd %esi, %xmm2 |
| ; KNL-NEXT: movl $274, %esi ## imm = 0x112 |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $29, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB22_28 |
| ; KNL-NEXT: ## BB#27: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB22_28: |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: vpinsrb $1, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %esi, %edi, %eax |
| ; KNL-NEXT: movzbl %r11b, %esi |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $30, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB22_30 |
| ; KNL-NEXT: ## BB#29: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB22_30: |
| ; KNL-NEXT: vpinsrb $2, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: movl $275, %eax ## imm = 0x113 |
| ; KNL-NEXT: bextrl %eax, %edi, %r11d |
| ; KNL-NEXT: movzbl %dl, %edx |
| ; KNL-NEXT: vmovd %esi, %xmm3 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $31, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_32 |
| ; KNL-NEXT: ## BB#31: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_32: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: vpinsrb $3, %r11d, %xmm2, %xmm2 |
| ; KNL-NEXT: movl $276, %eax ## imm = 0x114 |
| ; KNL-NEXT: bextrl %eax, %edi, %esi |
| ; KNL-NEXT: movl $277, %r11d ## imm = 0x115 |
| ; KNL-NEXT: vpinsrb $1, %edx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r10b, %r10d |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_34 |
| ; KNL-NEXT: ## BB#33: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_34: |
| ; KNL-NEXT: vpinsrb $4, %esi, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r11d, %edi, %edx |
| ; KNL-NEXT: movl $278, %r11d ## imm = 0x116 |
| ; KNL-NEXT: vpinsrb $2, %r10d, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r9b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shlq $63, %rcx |
| ; KNL-NEXT: sarq $63, %rcx |
| ; KNL-NEXT: vmovd %ecx, %xmm4 |
| ; KNL-NEXT: vpinsrb $1, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $2, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_36 |
| ; KNL-NEXT: ## BB#35: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_36: |
| ; KNL-NEXT: vpinsrb $5, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r11d, %edi, %edx |
| ; KNL-NEXT: movl $279, %r9d ## imm = 0x117 |
| ; KNL-NEXT: vpinsrb $3, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %bl, %ebx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $2, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $3, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_38 |
| ; KNL-NEXT: ## BB#37: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_38: |
| ; KNL-NEXT: vpinsrb $6, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r9d, %edi, %edx |
| ; KNL-NEXT: movl $280, %esi ## imm = 0x118 |
| ; KNL-NEXT: vpinsrb $4, %ebx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r12b, %ebx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $3, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $4, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_40 |
| ; KNL-NEXT: ## BB#39: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_40: |
| ; KNL-NEXT: vpinsrb $7, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %esi, %edi, %ecx |
| ; KNL-NEXT: movl $281, %edx ## imm = 0x119 |
| ; KNL-NEXT: vpinsrb $5, %ebx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r14b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $4, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $5, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_42 |
| ; KNL-NEXT: ## BB#41: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_42: |
| ; KNL-NEXT: vpinsrb $8, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $282, %edx ## imm = 0x11A |
| ; KNL-NEXT: vpinsrb $6, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r8b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $5, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %bl |
| ; KNL-NEXT: shrb $6, %bl |
| ; KNL-NEXT: andb $1, %bl |
| ; KNL-NEXT: je LBB22_44 |
| ; KNL-NEXT: ## BB#43: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB22_44: |
| ; KNL-NEXT: vpinsrb $9, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %eax |
| ; KNL-NEXT: movl $283, %ecx ## imm = 0x11B |
| ; KNL-NEXT: vpinsrb $7, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r13b, %esi |
| ; KNL-NEXT: movzbl %bl, %edx |
| ; KNL-NEXT: vpinsrb $6, %edx, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %bl |
| ; KNL-NEXT: shrb $7, %bl |
| ; KNL-NEXT: je LBB22_46 |
| ; KNL-NEXT: ## BB#45: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB22_46: |
| ; KNL-NEXT: vpinsrb $10, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: movl $284, %edx ## imm = 0x11C |
| ; KNL-NEXT: vpinsrb $8, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rax ## 8-byte Reload |
| ; KNL-NEXT: movzbl %al, %esi |
| ; KNL-NEXT: movzbl %bl, %eax |
| ; KNL-NEXT: vpinsrb $7, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $8, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_48 |
| ; KNL-NEXT: ## BB#47: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_48: |
| ; KNL-NEXT: vpinsrb $11, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $285, %edx ## imm = 0x11D |
| ; KNL-NEXT: vpinsrb $9, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rsi ## 8-byte Reload |
| ; KNL-NEXT: movzbl %sil, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $8, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $9, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_50 |
| ; KNL-NEXT: ## BB#49: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_50: |
| ; KNL-NEXT: vpinsrb $12, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $286, %edx ## imm = 0x11E |
| ; KNL-NEXT: vpinsrb $10, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rsi ## 8-byte Reload |
| ; KNL-NEXT: movzbl %sil, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $9, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $10, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_52 |
| ; KNL-NEXT: ## BB#51: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_52: |
| ; KNL-NEXT: vpinsrb $13, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %edx |
| ; KNL-NEXT: vpinsrb $11, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $10, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $11, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_54 |
| ; KNL-NEXT: ## BB#53: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_54: |
| ; KNL-NEXT: vpinsrb $14, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: shrl $31, %edi |
| ; KNL-NEXT: vpinsrb $12, %ecx, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $11, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $12, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_56 |
| ; KNL-NEXT: ## BB#55: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_56: |
| ; KNL-NEXT: vpinsrb $15, %edi, %xmm2, %xmm2 |
| ; KNL-NEXT: vpinsrb $13, %ecx, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $12, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $13, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_58 |
| ; KNL-NEXT: ## BB#57: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_58: |
| ; KNL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1 |
| ; KNL-NEXT: vpinsrb $14, %ecx, %xmm3, %xmm2 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $13, %eax, %xmm4, %xmm3 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $14, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB22_60 |
| ; KNL-NEXT: ## BB#59: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB22_60: |
| ; KNL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7] |
| ; KNL-NEXT: vpinsrb $15, %ecx, %xmm2, %xmm1 |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $14, %eax, %xmm3, %xmm2 |
| ; KNL-NEXT: shrq $15, %r15 |
| ; KNL-NEXT: andb $1, %r15b |
| ; KNL-NEXT: je LBB22_62 |
| ; KNL-NEXT: ## BB#61: |
| ; KNL-NEXT: movb $-1, %r15b |
| ; KNL-NEXT: LBB22_62: |
| ; KNL-NEXT: movzbl %r15b, %eax |
| ; KNL-NEXT: vpinsrb $15, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; KNL-NEXT: vpsllw $7, %ymm0, %ymm0 |
| ; KNL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 |
| ; KNL-NEXT: vpxor %ymm2, %ymm2, %ymm2 |
| ; KNL-NEXT: vpcmpgtb %ymm0, %ymm2, %ymm0 |
| ; KNL-NEXT: leaq -40(%rbp), %rsp |
| ; KNL-NEXT: popq %rbx |
| ; KNL-NEXT: popq %r12 |
| ; KNL-NEXT: popq %r13 |
| ; KNL-NEXT: popq %r14 |
| ; KNL-NEXT: popq %r15 |
| ; KNL-NEXT: popq %rbp |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test16: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovq %rdi, %k0 |
| ; SKX-NEXT: kxnorw %k0, %k0, %k1 |
| ; SKX-NEXT: kshiftrw $15, %k1, %k1 |
| ; SKX-NEXT: kshiftlq $5, %k1, %k1 |
| ; SKX-NEXT: korq %k1, %k0, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %zmm0 |
| ; SKX-NEXT: retq |
| %a = bitcast i64 %x to <64 x i1> |
| %b = insertelement <64 x i1>%a, i1 true, i32 5 |
| %c = sext <64 x i1>%b to <64 x i8> |
| ret <64 x i8>%c |
| } |
| |
| define <64 x i8> @test17(i64 %x, i32 %y, i32 %z) { |
| ; KNL-LABEL: test17: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: pushq %rbp |
| ; KNL-NEXT: Ltmp8: |
| ; KNL-NEXT: .cfi_def_cfa_offset 16 |
| ; KNL-NEXT: Ltmp9: |
| ; KNL-NEXT: .cfi_offset %rbp, -16 |
| ; KNL-NEXT: movq %rsp, %rbp |
| ; KNL-NEXT: Ltmp10: |
| ; KNL-NEXT: .cfi_def_cfa_register %rbp |
| ; KNL-NEXT: pushq %r15 |
| ; KNL-NEXT: pushq %r14 |
| ; KNL-NEXT: pushq %r13 |
| ; KNL-NEXT: pushq %r12 |
| ; KNL-NEXT: pushq %rbx |
| ; KNL-NEXT: andq $-32, %rsp |
| ; KNL-NEXT: subq $128, %rsp |
| ; KNL-NEXT: Ltmp11: |
| ; KNL-NEXT: .cfi_offset %rbx, -56 |
| ; KNL-NEXT: Ltmp12: |
| ; KNL-NEXT: .cfi_offset %r12, -48 |
| ; KNL-NEXT: Ltmp13: |
| ; KNL-NEXT: .cfi_offset %r13, -40 |
| ; KNL-NEXT: Ltmp14: |
| ; KNL-NEXT: .cfi_offset %r14, -32 |
| ; KNL-NEXT: Ltmp15: |
| ; KNL-NEXT: .cfi_offset %r15, -24 |
| ; KNL-NEXT: movq %rdi, %rax |
| ; KNL-NEXT: shrq $32, %rax |
| ; KNL-NEXT: movl %eax, {{[0-9]+}}(%rsp) |
| ; KNL-NEXT: movl %edi, %eax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: vmovd %eax, %xmm0 |
| ; KNL-NEXT: movl $257, %eax ## imm = 0x101 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $1, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $258, %eax ## imm = 0x102 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $2, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $259, %eax ## imm = 0x103 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $3, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $260, %eax ## imm = 0x104 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $4, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $261, %eax ## imm = 0x105 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $5, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $262, %eax ## imm = 0x106 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $6, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $263, %eax ## imm = 0x107 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $7, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $264, %eax ## imm = 0x108 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $8, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $265, %eax ## imm = 0x109 |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $9, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $266, %eax ## imm = 0x10A |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $10, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $267, %eax ## imm = 0x10B |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $11, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $268, %eax ## imm = 0x10C |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $12, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $269, %eax ## imm = 0x10D |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $13, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $270, %eax ## imm = 0x10E |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $14, %eax, %xmm0, %xmm0 |
| ; KNL-NEXT: movl $271, %eax ## imm = 0x10F |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vpinsrb $15, %eax, %xmm0, %xmm1 |
| ; KNL-NEXT: cmpl %edx, %esi |
| ; KNL-NEXT: setg %al |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $5, %eax, %xmm1, %xmm0 |
| ; KNL-NEXT: movl {{[0-9]+}}(%rsp), %r15d |
| ; KNL-NEXT: movq %r15, %rdx |
| ; KNL-NEXT: shrq $17, %rdx |
| ; KNL-NEXT: andb $1, %dl |
| ; KNL-NEXT: je LBB23_2 |
| ; KNL-NEXT: ## BB#1: |
| ; KNL-NEXT: movb $-1, %dl |
| ; KNL-NEXT: LBB23_2: |
| ; KNL-NEXT: movq %r15, %r11 |
| ; KNL-NEXT: shrq $16, %r11 |
| ; KNL-NEXT: andb $1, %r11b |
| ; KNL-NEXT: je LBB23_4 |
| ; KNL-NEXT: ## BB#3: |
| ; KNL-NEXT: movb $-1, %r11b |
| ; KNL-NEXT: LBB23_4: |
| ; KNL-NEXT: movq %r15, %r10 |
| ; KNL-NEXT: shrq $18, %r10 |
| ; KNL-NEXT: andb $1, %r10b |
| ; KNL-NEXT: je LBB23_6 |
| ; KNL-NEXT: ## BB#5: |
| ; KNL-NEXT: movb $-1, %r10b |
| ; KNL-NEXT: LBB23_6: |
| ; KNL-NEXT: movq %r15, %r9 |
| ; KNL-NEXT: shrq $19, %r9 |
| ; KNL-NEXT: andb $1, %r9b |
| ; KNL-NEXT: je LBB23_8 |
| ; KNL-NEXT: ## BB#7: |
| ; KNL-NEXT: movb $-1, %r9b |
| ; KNL-NEXT: LBB23_8: |
| ; KNL-NEXT: movq %r15, %rbx |
| ; KNL-NEXT: shrq $20, %rbx |
| ; KNL-NEXT: andb $1, %bl |
| ; KNL-NEXT: je LBB23_10 |
| ; KNL-NEXT: ## BB#9: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB23_10: |
| ; KNL-NEXT: movq %r15, %r12 |
| ; KNL-NEXT: shrq $21, %r12 |
| ; KNL-NEXT: andb $1, %r12b |
| ; KNL-NEXT: je LBB23_12 |
| ; KNL-NEXT: ## BB#11: |
| ; KNL-NEXT: movb $-1, %r12b |
| ; KNL-NEXT: LBB23_12: |
| ; KNL-NEXT: movq %r15, %r14 |
| ; KNL-NEXT: shrq $22, %r14 |
| ; KNL-NEXT: andb $1, %r14b |
| ; KNL-NEXT: je LBB23_14 |
| ; KNL-NEXT: ## BB#13: |
| ; KNL-NEXT: movb $-1, %r14b |
| ; KNL-NEXT: LBB23_14: |
| ; KNL-NEXT: movq %r15, %r8 |
| ; KNL-NEXT: shrq $23, %r8 |
| ; KNL-NEXT: andb $1, %r8b |
| ; KNL-NEXT: je LBB23_16 |
| ; KNL-NEXT: ## BB#15: |
| ; KNL-NEXT: movb $-1, %r8b |
| ; KNL-NEXT: LBB23_16: |
| ; KNL-NEXT: movq %r15, %r13 |
| ; KNL-NEXT: shrq $24, %r13 |
| ; KNL-NEXT: andb $1, %r13b |
| ; KNL-NEXT: je LBB23_18 |
| ; KNL-NEXT: ## BB#17: |
| ; KNL-NEXT: movb $-1, %r13b |
| ; KNL-NEXT: LBB23_18: |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $25, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_20 |
| ; KNL-NEXT: ## BB#19: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_20: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $26, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_22 |
| ; KNL-NEXT: ## BB#21: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_22: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movl $272, %esi ## imm = 0x110 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $27, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_24 |
| ; KNL-NEXT: ## BB#23: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_24: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movl $273, %eax ## imm = 0x111 |
| ; KNL-NEXT: bextrl %esi, %edi, %esi |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $28, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB23_26 |
| ; KNL-NEXT: ## BB#25: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB23_26: |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: bextrl %eax, %edi, %eax |
| ; KNL-NEXT: vmovd %esi, %xmm2 |
| ; KNL-NEXT: movl $274, %esi ## imm = 0x112 |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $29, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB23_28 |
| ; KNL-NEXT: ## BB#27: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB23_28: |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: vpinsrb $1, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %esi, %edi, %eax |
| ; KNL-NEXT: movzbl %r11b, %esi |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shrq $30, %rcx |
| ; KNL-NEXT: andb $1, %cl |
| ; KNL-NEXT: je LBB23_30 |
| ; KNL-NEXT: ## BB#29: |
| ; KNL-NEXT: movb $-1, %cl |
| ; KNL-NEXT: LBB23_30: |
| ; KNL-NEXT: vpinsrb $2, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: movl $275, %eax ## imm = 0x113 |
| ; KNL-NEXT: bextrl %eax, %edi, %r11d |
| ; KNL-NEXT: movzbl %dl, %edx |
| ; KNL-NEXT: vmovd %esi, %xmm3 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $31, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_32 |
| ; KNL-NEXT: ## BB#31: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_32: |
| ; KNL-NEXT: movq %rax, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: movq %rcx, {{[0-9]+}}(%rsp) ## 8-byte Spill |
| ; KNL-NEXT: vpinsrb $3, %r11d, %xmm2, %xmm2 |
| ; KNL-NEXT: movl $276, %eax ## imm = 0x114 |
| ; KNL-NEXT: bextrl %eax, %edi, %esi |
| ; KNL-NEXT: movl $277, %r11d ## imm = 0x115 |
| ; KNL-NEXT: vpinsrb $1, %edx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r10b, %r10d |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_34 |
| ; KNL-NEXT: ## BB#33: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_34: |
| ; KNL-NEXT: vpinsrb $4, %esi, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r11d, %edi, %edx |
| ; KNL-NEXT: movl $278, %r11d ## imm = 0x116 |
| ; KNL-NEXT: vpinsrb $2, %r10d, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r9b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: movq %r15, %rcx |
| ; KNL-NEXT: shlq $63, %rcx |
| ; KNL-NEXT: sarq $63, %rcx |
| ; KNL-NEXT: vmovd %ecx, %xmm4 |
| ; KNL-NEXT: vpinsrb $1, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $2, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_36 |
| ; KNL-NEXT: ## BB#35: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_36: |
| ; KNL-NEXT: vpinsrb $5, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r11d, %edi, %edx |
| ; KNL-NEXT: movl $279, %r9d ## imm = 0x117 |
| ; KNL-NEXT: vpinsrb $3, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %bl, %ebx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $2, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $3, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_38 |
| ; KNL-NEXT: ## BB#37: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_38: |
| ; KNL-NEXT: vpinsrb $6, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %r9d, %edi, %edx |
| ; KNL-NEXT: movl $280, %esi ## imm = 0x118 |
| ; KNL-NEXT: vpinsrb $4, %ebx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r12b, %ebx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $3, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $4, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_40 |
| ; KNL-NEXT: ## BB#39: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_40: |
| ; KNL-NEXT: vpinsrb $7, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %esi, %edi, %ecx |
| ; KNL-NEXT: movl $281, %edx ## imm = 0x119 |
| ; KNL-NEXT: vpinsrb $5, %ebx, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r14b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $4, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %al |
| ; KNL-NEXT: shrb $5, %al |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_42 |
| ; KNL-NEXT: ## BB#41: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_42: |
| ; KNL-NEXT: vpinsrb $8, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $282, %edx ## imm = 0x11A |
| ; KNL-NEXT: vpinsrb $6, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r8b, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $5, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %bl |
| ; KNL-NEXT: shrb $6, %bl |
| ; KNL-NEXT: andb $1, %bl |
| ; KNL-NEXT: je LBB23_44 |
| ; KNL-NEXT: ## BB#43: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB23_44: |
| ; KNL-NEXT: vpinsrb $9, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %eax |
| ; KNL-NEXT: movl $283, %ecx ## imm = 0x11B |
| ; KNL-NEXT: vpinsrb $7, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movzbl %r13b, %esi |
| ; KNL-NEXT: movzbl %bl, %edx |
| ; KNL-NEXT: vpinsrb $6, %edx, %xmm4, %xmm4 |
| ; KNL-NEXT: movb %r15b, %bl |
| ; KNL-NEXT: shrb $7, %bl |
| ; KNL-NEXT: je LBB23_46 |
| ; KNL-NEXT: ## BB#45: |
| ; KNL-NEXT: movb $-1, %bl |
| ; KNL-NEXT: LBB23_46: |
| ; KNL-NEXT: vpinsrb $10, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %ecx, %edi, %ecx |
| ; KNL-NEXT: movl $284, %edx ## imm = 0x11C |
| ; KNL-NEXT: vpinsrb $8, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rax ## 8-byte Reload |
| ; KNL-NEXT: movzbl %al, %esi |
| ; KNL-NEXT: movzbl %bl, %eax |
| ; KNL-NEXT: vpinsrb $7, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $8, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_48 |
| ; KNL-NEXT: ## BB#47: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_48: |
| ; KNL-NEXT: vpinsrb $11, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $285, %edx ## imm = 0x11D |
| ; KNL-NEXT: vpinsrb $9, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rsi ## 8-byte Reload |
| ; KNL-NEXT: movzbl %sil, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $8, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $9, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_50 |
| ; KNL-NEXT: ## BB#49: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_50: |
| ; KNL-NEXT: vpinsrb $12, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %ecx |
| ; KNL-NEXT: movl $286, %edx ## imm = 0x11E |
| ; KNL-NEXT: vpinsrb $10, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rsi ## 8-byte Reload |
| ; KNL-NEXT: movzbl %sil, %esi |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $9, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $10, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_52 |
| ; KNL-NEXT: ## BB#51: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_52: |
| ; KNL-NEXT: vpinsrb $13, %ecx, %xmm2, %xmm2 |
| ; KNL-NEXT: bextrl %edx, %edi, %edx |
| ; KNL-NEXT: vpinsrb $11, %esi, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $10, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $11, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_54 |
| ; KNL-NEXT: ## BB#53: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_54: |
| ; KNL-NEXT: vpinsrb $14, %edx, %xmm2, %xmm2 |
| ; KNL-NEXT: shrl $31, %edi |
| ; KNL-NEXT: vpinsrb $12, %ecx, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $11, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $12, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_56 |
| ; KNL-NEXT: ## BB#55: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_56: |
| ; KNL-NEXT: vpinsrb $15, %edi, %xmm2, %xmm2 |
| ; KNL-NEXT: vpinsrb $13, %ecx, %xmm3, %xmm3 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $12, %eax, %xmm4, %xmm4 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $13, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_58 |
| ; KNL-NEXT: ## BB#57: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_58: |
| ; KNL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1 |
| ; KNL-NEXT: vpinsrb $14, %ecx, %xmm3, %xmm2 |
| ; KNL-NEXT: movq {{[0-9]+}}(%rsp), %rcx ## 8-byte Reload |
| ; KNL-NEXT: movzbl %cl, %ecx |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $13, %eax, %xmm4, %xmm3 |
| ; KNL-NEXT: movq %r15, %rax |
| ; KNL-NEXT: shrq $14, %rax |
| ; KNL-NEXT: andb $1, %al |
| ; KNL-NEXT: je LBB23_60 |
| ; KNL-NEXT: ## BB#59: |
| ; KNL-NEXT: movb $-1, %al |
| ; KNL-NEXT: LBB23_60: |
| ; KNL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7] |
| ; KNL-NEXT: vpinsrb $15, %ecx, %xmm2, %xmm1 |
| ; KNL-NEXT: movzbl %al, %eax |
| ; KNL-NEXT: vpinsrb $14, %eax, %xmm3, %xmm2 |
| ; KNL-NEXT: shrq $15, %r15 |
| ; KNL-NEXT: andb $1, %r15b |
| ; KNL-NEXT: je LBB23_62 |
| ; KNL-NEXT: ## BB#61: |
| ; KNL-NEXT: movb $-1, %r15b |
| ; KNL-NEXT: LBB23_62: |
| ; KNL-NEXT: movzbl %r15b, %eax |
| ; KNL-NEXT: vpinsrb $15, %eax, %xmm2, %xmm2 |
| ; KNL-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; KNL-NEXT: vpsllw $7, %ymm0, %ymm0 |
| ; KNL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 |
| ; KNL-NEXT: vpxor %ymm2, %ymm2, %ymm2 |
| ; KNL-NEXT: vpcmpgtb %ymm0, %ymm2, %ymm0 |
| ; KNL-NEXT: leaq -40(%rbp), %rsp |
| ; KNL-NEXT: popq %rbx |
| ; KNL-NEXT: popq %r12 |
| ; KNL-NEXT: popq %r13 |
| ; KNL-NEXT: popq %r14 |
| ; KNL-NEXT: popq %r15 |
| ; KNL-NEXT: popq %rbp |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test17: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovq %rdi, %k0 |
| ; SKX-NEXT: cmpl %edx, %esi |
| ; SKX-NEXT: setg %al |
| ; SKX-NEXT: andl $1, %eax |
| ; SKX-NEXT: kmovw %eax, %k1 |
| ; SKX-NEXT: kshiftlq $5, %k1, %k1 |
| ; SKX-NEXT: korq %k1, %k0, %k0 |
| ; SKX-NEXT: vpmovm2b %k0, %zmm0 |
| ; SKX-NEXT: retq |
| %a = bitcast i64 %x to <64 x i1> |
| %b = icmp sgt i32 %y, %z |
| %c = insertelement <64 x i1>%a, i1 %b, i32 5 |
| %d = sext <64 x i1>%c to <64 x i8> |
| ret <64 x i8>%d |
| } |
| |
| define <8 x i1> @test18(i8 %a, i16 %y) { |
| ; KNL-LABEL: test18: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: movzbl %dil, %eax |
| ; KNL-NEXT: kmovw %eax, %k0 |
| ; KNL-NEXT: kmovw %esi, %k1 |
| ; KNL-NEXT: kshiftlw $7, %k1, %k2 |
| ; KNL-NEXT: kshiftrw $15, %k2, %k2 |
| ; KNL-NEXT: kshiftlw $6, %k1, %k1 |
| ; KNL-NEXT: kshiftrw $15, %k1, %k1 |
| ; KNL-NEXT: kshiftlw $6, %k1, %k1 |
| ; KNL-NEXT: korw %k1, %k0, %k0 |
| ; KNL-NEXT: kshiftlw $7, %k2, %k1 |
| ; KNL-NEXT: korw %k1, %k0, %k1 |
| ; KNL-NEXT: vpbroadcastq {{.*}}(%rip), %zmm0 {%k1} {z} |
| ; KNL-NEXT: vpmovqw %zmm0, %xmm0 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test18: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: kmovb %edi, %k0 |
| ; SKX-NEXT: kmovw %esi, %k1 |
| ; SKX-NEXT: kshiftlw $6, %k1, %k2 |
| ; SKX-NEXT: kshiftrw $15, %k2, %k2 |
| ; SKX-NEXT: kshiftlw $7, %k1, %k1 |
| ; SKX-NEXT: kshiftrw $15, %k1, %k1 |
| ; SKX-NEXT: kshiftlb $7, %k1, %k1 |
| ; SKX-NEXT: kshiftlb $6, %k2, %k2 |
| ; SKX-NEXT: korb %k2, %k0, %k0 |
| ; SKX-NEXT: korb %k1, %k0, %k0 |
| ; SKX-NEXT: vpmovm2w %k0, %xmm0 |
| ; SKX-NEXT: retq |
| %b = bitcast i8 %a to <8 x i1> |
| %b1 = bitcast i16 %y to <16 x i1> |
| %el1 = extractelement <16 x i1>%b1, i32 8 |
| %el2 = extractelement <16 x i1>%b1, i32 9 |
| %c = insertelement <8 x i1>%b, i1 %el1, i32 7 |
| %d = insertelement <8 x i1>%c, i1 %el2, i32 6 |
| ret <8 x i1>%d |
| } |
| define <32 x i16> @test21(<32 x i16> %x , <32 x i1> %mask) nounwind readnone { |
| ; KNL-LABEL: test21: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpmovzxbw {{.*#+}} ymm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero |
| ; KNL-NEXT: vpsllw $15, %ymm3, %ymm3 |
| ; KNL-NEXT: vpsraw $15, %ymm3, %ymm3 |
| ; KNL-NEXT: vpand %ymm0, %ymm3, %ymm0 |
| ; KNL-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; KNL-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero |
| ; KNL-NEXT: vpsllw $15, %ymm2, %ymm2 |
| ; KNL-NEXT: vpsraw $15, %ymm2, %ymm2 |
| ; KNL-NEXT: vpand %ymm1, %ymm2, %ymm1 |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test21: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpsllw $7, %ymm1, %ymm1 |
| ; SKX-NEXT: vpmovb2m %ymm1, %k1 |
| ; SKX-NEXT: vpxord %zmm1, %zmm1, %zmm1 |
| ; SKX-NEXT: vmovdqu16 %zmm0, %zmm1 {%k1} |
| ; SKX-NEXT: vmovaps %zmm1, %zmm0 |
| ; SKX-NEXT: retq |
| %ret = select <32 x i1> %mask, <32 x i16> %x, <32 x i16> zeroinitializer |
| ret <32 x i16> %ret |
| } |
| |
| define void @test22(<4 x i1> %a, <4 x i1>* %addr) { |
| ; KNL-LABEL: test22: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpextrd $3, %xmm0, %eax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: vpextrd $2, %xmm0, %eax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: vpextrd $1, %xmm0, %eax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: vmovd %xmm0, %eax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test22: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpslld $31, %xmm0, %xmm0 |
| ; SKX-NEXT: vpmovd2m %xmm0, %k0 |
| ; SKX-NEXT: kmovb %k0, (%rdi) |
| ; SKX-NEXT: retq |
| store <4 x i1> %a, <4 x i1>* %addr |
| ret void |
| } |
| |
| define void @test23(<2 x i1> %a, <2 x i1>* %addr) { |
| ; KNL-LABEL: test23: |
| ; KNL: ## BB#0: |
| ; KNL-NEXT: vpextrq $1, %xmm0, %rax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: vmovq %xmm0, %rax |
| ; KNL-NEXT: andl $1, %eax |
| ; KNL-NEXT: movb %al, (%rdi) |
| ; KNL-NEXT: retq |
| ; |
| ; SKX-LABEL: test23: |
| ; SKX: ## BB#0: |
| ; SKX-NEXT: vpsllq $63, %xmm0, %xmm0 |
| ; SKX-NEXT: vpmovq2m %xmm0, %k0 |
| ; SKX-NEXT: kmovb %k0, (%rdi) |
| ; SKX-NEXT: retq |
| store <2 x i1> %a, <2 x i1>* %addr |
| ret void |
| } |