AVX512: Change store size of kmask. Store size of v8i1, v4i1 , v2i1 and i1 are changed to 16 bits.

If KMOVB not supported (require AVX512DQ) only KMOVW can be used so store size should be 2 bytes.

Differential Revision: http://reviews.llvm.org/D17138

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@260878 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 130 insertions, 4 deletions

diff --git a/test/CodeGen/X86/avx512-intel-ocl.ll b/test/CodeGen/X86/avx512-intel-ocl.ll
index 2e1b27e4aec..70c86347fad 100644
--- a/test/CodeGen/X86/avx512-intel-ocl.ll
+++ b/test/CodeGen/X86/avx512-intel-ocl.ll
@@ -68,10 +68,10 @@ define <16 x float> @testf16_regs(<16 x float> %a, <16 x float> %b) nounwind {
 ; WIN64: vmovups {{.*(%rbp).*}}, %zmm21     # 64-byte Reload
 
 ; X64-LABEL: test_prolog_epilog
-; X64:  kmovw   %k7, {{.*}}(%rsp)         ## 8-byte Folded Spill
-; X64:  kmovw   %k6, {{.*}}(%rsp)         ## 8-byte Folded Spill
-; X64:  kmovw   %k5, {{.*}}(%rsp)         ## 8-byte Folded Spill
-; X64:  kmovw   %k4, {{.*}}(%rsp)         ## 8-byte Folded Spill
+; X64:  kmovq   %k7, {{.*}}(%rsp)         ## 8-byte Folded Spill
+; X64:  kmovq   %k6, {{.*}}(%rsp)         ## 8-byte Folded Spill
+; X64:  kmovq   %k5, {{.*}}(%rsp)         ## 8-byte Folded Spill
+; X64:  kmovq   %k4, {{.*}}(%rsp)         ## 8-byte Folded Spill
 ; X64: vmovups %zmm31, {{.*}}(%rsp)  ## 64-byte Spill
 ; X64: vmovups %zmm16, {{.*}}(%rsp)  ## 64-byte Spill
 ; X64: call
diff --git a/test/CodeGen/X86/avx512-mask-spills.ll b/test/CodeGen/X86/avx512-mask-spills.ll
new file mode 100644
index 00000000000..1be732a2e94
--- /dev/null
+++ b/test/CodeGen/X86/avx512-mask-spills.ll
@@ -0,0 +1,126 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
+
+declare void @f()
+define <4 x i1> @test_4i1(<4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: test_4i1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:  Ltmp0:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    vpcmpnleud %xmm1, %xmm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    vpcmpgtd %xmm1, %xmm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    callq _f
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k0 ## 2-byte Folded Reload
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1 ## 2-byte Folded Reload
+; CHECK-NEXT:    korw %k1, %k0, %k0
+; CHECK-NEXT:    vpmovm2d %k0, %xmm0
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+
+  %cmp_res = icmp ugt <4 x i32> %a, %b
+  %cmp_res2 = icmp sgt <4 x i32> %a, %b
+  call void @f()
+  %res = or <4 x i1> %cmp_res, %cmp_res2
+  ret <4 x i1> %res
+}
+
+define <8 x i1> @test_8i1(<8 x i32> %a, <8 x i32> %b) {
+; CHECK-LABEL: test_8i1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:  Ltmp1:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    vpcmpnleud %ymm1, %ymm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    vpcmpgtd %ymm1, %ymm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    callq _f
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k0 ## 2-byte Folded Reload
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1 ## 2-byte Folded Reload
+; CHECK-NEXT:    korb %k1, %k0, %k0
+; CHECK-NEXT:    vpmovm2w %k0, %xmm0
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+
+  %cmp_res = icmp ugt <8 x i32> %a, %b
+  %cmp_res2 = icmp sgt <8 x i32> %a, %b
+  call void @f()
+  %res = or <8 x i1> %cmp_res, %cmp_res2
+  ret <8 x i1> %res
+}
+
+define <16 x i1> @test_16i1(<16 x i32> %a, <16 x i32> %b) {
+; CHECK-LABEL: test_16i1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:  Ltmp2:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovw %k0, {{[0-9]+}}(%rsp) ## 2-byte Folded Spill
+; CHECK-NEXT:    callq _f
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k0 ## 2-byte Folded Reload
+; CHECK-NEXT:    kmovw {{[0-9]+}}(%rsp), %k1 ## 2-byte Folded Reload
+; CHECK-NEXT:    korw %k1, %k0, %k0
+; CHECK-NEXT:    vpmovm2b %k0, %xmm0
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+  %cmp_res = icmp ugt <16 x i32> %a, %b
+  %cmp_res2 = icmp sgt <16 x i32> %a, %b
+  call void @f()
+  %res = or <16 x i1> %cmp_res, %cmp_res2
+  ret <16 x i1> %res
+}
+
+define <32 x i1> @test_32i1(<32 x i16> %a, <32 x i16> %b) {
+; CHECK-LABEL: test_32i1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:  Ltmp3:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    vpcmpnleuw %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovd %k0, {{[0-9]+}}(%rsp) ## 4-byte Folded Spill
+; CHECK-NEXT:    vpcmpgtw %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovd %k0, (%rsp) ## 4-byte Folded Spill
+; CHECK-NEXT:    callq _f
+; CHECK-NEXT:    kmovd {{[0-9]+}}(%rsp), %k0 ## 4-byte Folded Reload
+; CHECK-NEXT:    kmovd (%rsp), %k1 ## 4-byte Folded Reload
+; CHECK-NEXT:    kord %k1, %k0, %k0
+; CHECK-NEXT:    vpmovm2b %k0, %ymm0
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+  %cmp_res = icmp ugt <32 x i16> %a, %b
+  %cmp_res2 = icmp sgt <32 x i16> %a, %b
+  call void @f()
+  %res = or <32 x i1> %cmp_res, %cmp_res2
+  ret <32 x i1> %res
+}
+
+define <64 x i1> @test_64i1(<64 x i8> %a, <64 x i8> %b) {
+; CHECK-LABEL: test_64i1:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    subq $24, %rsp
+; CHECK-NEXT:  Ltmp4:
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    vpcmpnleub %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovq %k0, {{[0-9]+}}(%rsp) ## 8-byte Folded Spill
+; CHECK-NEXT:    vpcmpgtb %zmm1, %zmm0, %k0
+; CHECK-NEXT:    kmovq %k0, {{[0-9]+}}(%rsp) ## 8-byte Folded Spill
+; CHECK-NEXT:    callq _f
+; CHECK-NEXT:    kmovq {{[0-9]+}}(%rsp), %k0 ## 8-byte Folded Reload
+; CHECK-NEXT:    kmovq {{[0-9]+}}(%rsp), %k1 ## 8-byte Folded Reload
+; CHECK-NEXT:    korq %k1, %k0, %k0
+; CHECK-NEXT:    vpmovm2b %k0, %zmm0
+; CHECK-NEXT:    addq $24, %rsp
+; CHECK-NEXT:    retq
+
+  %cmp_res = icmp ugt <64 x i8> %a, %b
+  %cmp_res2 = icmp sgt <64 x i8> %a, %b
+  call void @f()
+  %res = or <64 x i1> %cmp_res, %cmp_res2
+  ret <64 x i1> %res
+}