[AVX512] Added load/store instructions to Register2Memory opcode tables.

Added lowering tests for load/store.

Reviewed by Elena Demikhovsky <elena.demikhovsky@intel.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@214972 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 36ee4a2d26b0107a77cf80ae09cb9cd93e17aaa5..0d46f706906ee65d8832eeacfae12b342ba81ba4 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -377,7 +377,15 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::VMOVUPDYrr,  X86::VMOVUPDYmr,    TB_FOLDED_STORE },
     { X86::VMOVUPSYrr,  X86::VMOVUPSYmr,    TB_FOLDED_STORE },
     // AVX-512 foldable instructions
-    { X86::VMOVPDI2DIZrr,X86::VMOVPDI2DIZmr,  TB_FOLDED_STORE }
+    { X86::VMOVPDI2DIZrr,   X86::VMOVPDI2DIZmr, TB_FOLDED_STORE },
+    { X86::VMOVAPDZrr,      X86::VMOVAPDZmr,    TB_FOLDED_STORE | TB_ALIGN_64 },
+    { X86::VMOVAPSZrr,      X86::VMOVAPSZmr,    TB_FOLDED_STORE | TB_ALIGN_64 },
+    { X86::VMOVDQA32Zrr,    X86::VMOVDQA32Zmr,  TB_FOLDED_STORE | TB_ALIGN_64 },
+    { X86::VMOVDQA64Zrr,    X86::VMOVDQA64Zmr,  TB_FOLDED_STORE | TB_ALIGN_64 },
+    { X86::VMOVUPDZrr,      X86::VMOVUPDZmr,    TB_FOLDED_STORE },
+    { X86::VMOVUPSZrr,      X86::VMOVUPSZmr,    TB_FOLDED_STORE },
+    { X86::VMOVDQU32Zrr,    X86::VMOVDQU32Zmr,  TB_FOLDED_STORE },
+    { X86::VMOVDQU64Zrr,    X86::VMOVDQU64Zmr,  TB_FOLDED_STORE }
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl0); i != e; ++i) {
@@ -601,10 +609,14 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     // AVX-512 foldable instructions
     { X86::VMOV64toPQIZrr,  X86::VMOVQI2PQIZrm,       0 },
     { X86::VMOVDI2SSZrr,    X86::VMOVDI2SSZrm,        0 },
+    { X86::VMOVAPDZrr,      X86::VMOVAPDZrm,          TB_ALIGN_64 },
+    { X86::VMOVAPSZrr,      X86::VMOVAPSZrm,          TB_ALIGN_64 },
     { X86::VMOVDQA32Zrr,    X86::VMOVDQA32Zrm,        TB_ALIGN_64 },
     { X86::VMOVDQA64Zrr,    X86::VMOVDQA64Zrm,        TB_ALIGN_64 },
     { X86::VMOVDQU32Zrr,    X86::VMOVDQU32Zrm,        0 },
     { X86::VMOVDQU64Zrr,    X86::VMOVDQU64Zrm,        0 },
+    { X86::VMOVUPDZrr,      X86::VMOVUPDZrm,          0 },
+    { X86::VMOVUPSZrr,      X86::VMOVUPSZrm,          0 },
     { X86::VPABSDZrr,       X86::VPABSDZrm,           0 },
     { X86::VPABSQZrr,       X86::VPABSQZrm,           0 },
 
@@ -612,7 +624,7 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::AESIMCrr,              X86::AESIMCrm,              TB_ALIGN_16 },
     { X86::AESKEYGENASSIST128rr,  X86::AESKEYGENASSIST128rm,  TB_ALIGN_16 },
     { X86::VAESIMCrr,             X86::VAESIMCrm,             TB_ALIGN_16 },
-    { X86::VAESKEYGENASSIST128rr, X86::VAESKEYGENASSIST128rm, TB_ALIGN_16 },
+    { X86::VAESKEYGENASSIST128rr, X86::VAESKEYGENASSIST128rm, TB_ALIGN_16 }
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {

diff --git a/test/CodeGen/X86/avx512-mov.ll b/test/CodeGen/X86/avx512-mov.ll
index 6d80c994e100977ae624b10650f04529722500ce..93875e839e2259166b854beefe2ff2e0888b67fb 100644 (file)
--- a/test/CodeGen/X86/avx512-mov.ll
+++ b/test/CodeGen/X86/avx512-mov.ll
@@ -152,3 +152,296 @@ define void @test18(i8 * %addr, <8 x i64> %data) {
   store <8 x i64>%data, <8 x i64>* %vaddr, align 64
   ret void
 }
+
+; CHECK-LABEL: test19
+; CHECK: vmovdqu32
+; CHECK: ret
+define void @test19(i8 * %addr, <16 x i32> %data) {
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  store <16 x i32>%data, <16 x i32>* %vaddr, align 1
+  ret void
+}
+
+; CHECK-LABEL: test20
+; CHECK: vmovdqa32
+; CHECK: ret
+define void @test20(i8 * %addr, <16 x i32> %data) {
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  store <16 x i32>%data, <16 x i32>* %vaddr, align 64
+  ret void
+}
+
+; CHECK-LABEL: test21
+; CHECK: vmovdqa64
+; CHECK: ret
+define  <8 x i64> @test21(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %res = load <8 x i64>* %vaddr, align 64
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test22
+; CHECK: vmovdqu64
+; CHECK: ret
+define void @test22(i8 * %addr, <8 x i64> %data) {
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  store <8 x i64>%data, <8 x i64>* %vaddr, align 1
+  ret void
+}
+
+; CHECK-LABEL: test23
+; CHECK: vmovdqu64
+; CHECK: ret
+define <8 x i64> @test23(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %res = load <8 x i64>* %vaddr, align 1
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test24
+; CHECK: vmovapd
+; CHECK: ret
+define void @test24(i8 * %addr, <8 x double> %data) {
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  store <8 x double>%data, <8 x double>* %vaddr, align 64
+  ret void
+}
+
+; CHECK-LABEL: test25
+; CHECK: vmovapd
+; CHECK: ret
+define <8 x double> @test25(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %res = load <8 x double>* %vaddr, align 64
+  ret <8 x double>%res
+}
+
+; CHECK-LABEL: test26
+; CHECK: vmovaps
+; CHECK: ret
+define void @test26(i8 * %addr, <16 x float> %data) {
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  store <16 x float>%data, <16 x float>* %vaddr, align 64
+  ret void
+}
+
+; CHECK-LABEL: test27
+; CHECK: vmovaps
+; CHECK: ret
+define <16 x float> @test27(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %res = load <16 x float>* %vaddr, align 64
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test28
+; CHECK: vmovupd
+; CHECK: ret
+define void @test28(i8 * %addr, <8 x double> %data) {
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  store <8 x double>%data, <8 x double>* %vaddr, align 1
+  ret void
+}
+
+; CHECK-LABEL: test29
+; CHECK: vmovupd
+; CHECK: ret
+define <8 x double> @test29(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %res = load <8 x double>* %vaddr, align 1
+  ret <8 x double>%res
+}
+
+; CHECK-LABEL: test30
+; CHECK: vmovups
+; CHECK: ret
+define void @test30(i8 * %addr, <16 x float> %data) {
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  store <16 x float>%data, <16 x float>* %vaddr, align 1
+  ret void
+}
+
+; CHECK-LABEL: test31
+; CHECK: vmovups
+; CHECK: ret
+define <16 x float> @test31(i8 * %addr) {
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %res = load <16 x float>* %vaddr, align 1
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test32
+; CHECK: vmovdqa32{{.*{%k[1-7]} }}
+; CHECK: ret
+define <16 x i32> @test32(i8 * %addr, <16 x i32> %old, <16 x i32> %mask1) {
+  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  %r = load <16 x i32>* %vaddr, align 64
+  %res = select <16 x i1> %mask, <16 x i32> %r, <16 x i32> %old
+  ret <16 x i32>%res
+}
+
+; CHECK-LABEL: test33
+; CHECK: vmovdqu32{{.*{%k[1-7]} }}
+; CHECK: ret
+define <16 x i32> @test33(i8 * %addr, <16 x i32> %old, <16 x i32> %mask1) {
+  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  %r = load <16 x i32>* %vaddr, align 1
+  %res = select <16 x i1> %mask, <16 x i32> %r, <16 x i32> %old
+  ret <16 x i32>%res
+}
+
+; CHECK-LABEL: test34
+; CHECK: vmovdqa32{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <16 x i32> @test34(i8 * %addr, <16 x i32> %mask1) {
+  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  %r = load <16 x i32>* %vaddr, align 64
+  %res = select <16 x i1> %mask, <16 x i32> %r, <16 x i32> zeroinitializer
+  ret <16 x i32>%res
+}
+
+; CHECK-LABEL: test35
+; CHECK: vmovdqu32{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <16 x i32> @test35(i8 * %addr, <16 x i32> %mask1) {
+  %mask = icmp ne <16 x i32> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x i32>*
+  %r = load <16 x i32>* %vaddr, align 1
+  %res = select <16 x i1> %mask, <16 x i32> %r, <16 x i32> zeroinitializer
+  ret <16 x i32>%res
+}
+
+; CHECK-LABEL: test36
+; CHECK: vmovdqa64{{.*{%k[1-7]} }}
+; CHECK: ret
+define <8 x i64> @test36(i8 * %addr, <8 x i64> %old, <8 x i64> %mask1) {
+  %mask = icmp ne <8 x i64> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %r = load <8 x i64>* %vaddr, align 64
+  %res = select <8 x i1> %mask, <8 x i64> %r, <8 x i64> %old
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test37
+; CHECK: vmovdqu64{{.*{%k[1-7]} }}
+; CHECK: ret
+define <8 x i64> @test37(i8 * %addr, <8 x i64> %old, <8 x i64> %mask1) {
+  %mask = icmp ne <8 x i64> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %r = load <8 x i64>* %vaddr, align 1
+  %res = select <8 x i1> %mask, <8 x i64> %r, <8 x i64> %old
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test38
+; CHECK: vmovdqa64{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <8 x i64> @test38(i8 * %addr, <8 x i64> %mask1) {
+  %mask = icmp ne <8 x i64> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %r = load <8 x i64>* %vaddr, align 64
+  %res = select <8 x i1> %mask, <8 x i64> %r, <8 x i64> zeroinitializer
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test39
+; CHECK: vmovdqu64{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <8 x i64> @test39(i8 * %addr, <8 x i64> %mask1) {
+  %mask = icmp ne <8 x i64> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x i64>*
+  %r = load <8 x i64>* %vaddr, align 1
+  %res = select <8 x i1> %mask, <8 x i64> %r, <8 x i64> zeroinitializer
+  ret <8 x i64>%res
+}
+
+; CHECK-LABEL: test40
+; CHECK: vmovaps{{.*{%k[1-7]} }}
+; CHECK: ret
+define <16 x float> @test40(i8 * %addr, <16 x float> %old, <16 x float> %mask1) {
+  %mask = fcmp one <16 x float> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %r = load <16 x float>* %vaddr, align 64
+  %res = select <16 x i1> %mask, <16 x float> %r, <16 x float> %old
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test41
+; CHECK: vmovups{{.*{%k[1-7]} }}
+; CHECK: ret
+define <16 x float> @test41(i8 * %addr, <16 x float> %old, <16 x float> %mask1) {
+  %mask = fcmp one <16 x float> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %r = load <16 x float>* %vaddr, align 1
+  %res = select <16 x i1> %mask, <16 x float> %r, <16 x float> %old
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test42
+; CHECK: vmovaps{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <16 x float> @test42(i8 * %addr, <16 x float> %mask1) {
+  %mask = fcmp one <16 x float> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %r = load <16 x float>* %vaddr, align 64
+  %res = select <16 x i1> %mask, <16 x float> %r, <16 x float> zeroinitializer
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test43
+; CHECK: vmovups{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <16 x float> @test43(i8 * %addr, <16 x float> %mask1) {
+  %mask = fcmp one <16 x float> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <16 x float>*
+  %r = load <16 x float>* %vaddr, align 1
+  %res = select <16 x i1> %mask, <16 x float> %r, <16 x float> zeroinitializer
+  ret <16 x float>%res
+}
+
+; CHECK-LABEL: test44
+; CHECK: vmovapd{{.*{%k[1-7]} }}
+; CHECK: ret
+define <8 x double> @test44(i8 * %addr, <8 x double> %old, <8 x double> %mask1) {
+  %mask = fcmp one <8 x double> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %r = load <8 x double>* %vaddr, align 64
+  %res = select <8 x i1> %mask, <8 x double> %r, <8 x double> %old
+  ret <8 x double>%res
+}
+
+; CHECK-LABEL: test45
+; CHECK: vmovupd{{.*{%k[1-7]} }}
+; CHECK: ret
+define <8 x double> @test45(i8 * %addr, <8 x double> %old, <8 x double> %mask1) {
+  %mask = fcmp one <8 x double> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %r = load <8 x double>* %vaddr, align 1
+  %res = select <8 x i1> %mask, <8 x double> %r, <8 x double> %old
+  ret <8 x double>%res
+}
+
+; CHECK-LABEL: test46
+; CHECK: vmovapd{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <8 x double> @test46(i8 * %addr, <8 x double> %mask1) {
+  %mask = fcmp one <8 x double> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %r = load <8 x double>* %vaddr, align 64
+  %res = select <8 x i1> %mask, <8 x double> %r, <8 x double> zeroinitializer
+  ret <8 x double>%res
+}
+
+; CHECK-LABEL: test47
+; CHECK: vmovupd{{.*{%k[1-7]} {z} }}
+; CHECK: ret
+define <8 x double> @test47(i8 * %addr, <8 x double> %mask1) {
+  %mask = fcmp one <8 x double> %mask1, zeroinitializer
+  %vaddr = bitcast i8* %addr to <8 x double>*
+  %r = load <8 x double>* %vaddr, align 1
+  %res = select <8 x i1> %mask, <8 x double> %r, <8 x double> zeroinitializer
+  ret <8 x double>%res
+}