2 files changed, 139 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
new file mode 100644
index 00000000000..06cc91452a7
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -0,0 +1,70 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw -mattr=+avx512vl --show-mc-encoding| FileCheck %s
+
+; 256-bit
+
+define i32 @test_pcmpeq_b_256(<32 x i8> %a, <32 x i8> %b) {
+; CHECK-LABEL: test_pcmpeq_b_256
+; CHECK: vpcmpeqb %ymm1, %ymm0, %k0 ##
+  %res = call i32 @llvm.x86.avx512.mask.pcmpeq.b.256(<32 x i8> %a, <32 x i8> %b, i32 -1)
+  ret i32 %res
+}
+
+define i32 @test_mask_pcmpeq_b_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_b_256
+; CHECK: vpcmpeqb %ymm1, %ymm0, %k0 {%k1} ##
+  %res = call i32 @llvm.x86.avx512.mask.pcmpeq.b.256(<32 x i8> %a, <32 x i8> %b, i32 %mask)
+  ret i32 %res
+}
+
+declare i32 @llvm.x86.avx512.mask.pcmpeq.b.256(<32 x i8>, <32 x i8>, i32)
+
+define i16 @test_pcmpeq_w_256(<16 x i16> %a, <16 x i16> %b) {
+; CHECK-LABEL: test_pcmpeq_w_256
+; CHECK: vpcmpeqw %ymm1, %ymm0, %k0 ##
+  %res = call i16 @llvm.x86.avx512.mask.pcmpeq.w.256(<16 x i16> %a, <16 x i16> %b, i16 -1)
+  ret i16 %res
+}
+
+define i16 @test_mask_pcmpeq_w_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_w_256
+; CHECK: vpcmpeqw %ymm1, %ymm0, %k0 {%k1} ##
+  %res = call i16 @llvm.x86.avx512.mask.pcmpeq.w.256(<16 x i16> %a, <16 x i16> %b, i16 %mask)
+  ret i16 %res
+}
+
+declare i16 @llvm.x86.avx512.mask.pcmpeq.w.256(<16 x i16>, <16 x i16>, i16)
+
+; 128-bit
+
+define i16 @test_pcmpeq_b_128(<16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: test_pcmpeq_b_128
+; CHECK: vpcmpeqb %xmm1, %xmm0, %k0 ##
+  %res = call i16 @llvm.x86.avx512.mask.pcmpeq.b.128(<16 x i8> %a, <16 x i8> %b, i16 -1)
+  ret i16 %res
+}
+
+define i16 @test_mask_pcmpeq_b_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_b_128
+; CHECK: vpcmpeqb %xmm1, %xmm0, %k0 {%k1} ##
+  %res = call i16 @llvm.x86.avx512.mask.pcmpeq.b.128(<16 x i8> %a, <16 x i8> %b, i16 %mask)
+  ret i16 %res
+}
+
+declare i16 @llvm.x86.avx512.mask.pcmpeq.b.128(<16 x i8>, <16 x i8>, i16)
+
+define i8 @test_pcmpeq_w_128(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: test_pcmpeq_w_128
+; CHECK: vpcmpeqw %xmm1, %xmm0, %k0 ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.w.128(<8 x i16> %a, <8 x i16> %b, i8 -1)
+  ret i8 %res
+}
+
+define i8 @test_mask_pcmpeq_w_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_w_128
+; CHECK: vpcmpeqw %xmm1, %xmm0, %k0 {%k1} ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.w.128(<8 x i16> %a, <8 x i16> %b, i8 %mask)
+  ret i8 %res
+}
+
+declare i8 @llvm.x86.avx512.mask.pcmpeq.w.128(<8 x i16>, <8 x i16>, i8)
+
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
new file mode 100644
index 00000000000..ada13dea249
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -0,0 +1,69 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512vl --show-mc-encoding| FileCheck %s
+
+; 256-bit
+
+define i8 @test_pcmpeq_d_256(<8 x i32> %a, <8 x i32> %b) {
+; CHECK-LABEL: test_pcmpeq_d_256
+; CHECK: vpcmpeqd %ymm1, %ymm0, %k0 ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.d.256(<8 x i32> %a, <8 x i32> %b, i8 -1)
+  ret i8 %res
+}
+
+define i8 @test_mask_pcmpeq_d_256(<8 x i32> %a, <8 x i32> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_d_256
+; CHECK: vpcmpeqd %ymm1, %ymm0, %k0 {%k1} ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.d.256(<8 x i32> %a, <8 x i32> %b, i8 %mask)
+  ret i8 %res
+}
+
+declare i8 @llvm.x86.avx512.mask.pcmpeq.d.256(<8 x i32>, <8 x i32>, i8)
+
+define i8 @test_pcmpeq_q_256(<4 x i64> %a, <4 x i64> %b) {
+; CHECK-LABEL: test_pcmpeq_q_256
+; CHECK: vpcmpeqq %ymm1, %ymm0, %k0 ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.q.256(<4 x i64> %a, <4 x i64> %b, i8 -1)
+  ret i8 %res
+}
+
+define i8 @test_mask_pcmpeq_q_256(<4 x i64> %a, <4 x i64> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_q_256
+; CHECK: vpcmpeqq %ymm1, %ymm0, %k0 {%k1} ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.q.256(<4 x i64> %a, <4 x i64> %b, i8 %mask)
+  ret i8 %res
+}
+
+declare i8 @llvm.x86.avx512.mask.pcmpeq.q.256(<4 x i64>, <4 x i64>, i8)
+
+; 128-bit
+
+define i8 @test_pcmpeq_d_128(<4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: test_pcmpeq_d_128
+; CHECK: vpcmpeqd %xmm1, %xmm0, %k0 ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.d.128(<4 x i32> %a, <4 x i32> %b, i8 -1)
+  ret i8 %res
+}
+
+define i8 @test_mask_pcmpeq_d_128(<4 x i32> %a, <4 x i32> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_d_128
+; CHECK: vpcmpeqd %xmm1, %xmm0, %k0 {%k1} ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.d.128(<4 x i32> %a, <4 x i32> %b, i8 %mask)
+  ret i8 %res
+}
+
+declare i8 @llvm.x86.avx512.mask.pcmpeq.d.128(<4 x i32>, <4 x i32>, i8)
+
+define i8 @test_pcmpeq_q_128(<2 x i64> %a, <2 x i64> %b) {
+; CHECK-LABEL: test_pcmpeq_q_128
+; CHECK: vpcmpeqq %xmm1, %xmm0, %k0 ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.q.128(<2 x i64> %a, <2 x i64> %b, i8 -1)
+  ret i8 %res
+}
+
+define i8 @test_mask_pcmpeq_q_128(<2 x i64> %a, <2 x i64> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_pcmpeq_q_128
+; CHECK: vpcmpeqq %xmm1, %xmm0, %k0 {%k1} ##
+  %res = call i8 @llvm.x86.avx512.mask.pcmpeq.q.128(<2 x i64> %a, <2 x i64> %b, i8 %mask)
+  ret i8 %res
+}
+
+declare i8 @llvm.x86.avx512.mask.pcmpeq.q.128(<2 x i64>, <2 x i64>, i8)