[X86] Use sse_load_f32/f64 to improve load folding for scalar VFPCLASS intrinsics.

llvm-svn: 318019

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 12b3787..5a59645 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -2365,18 +2365,18 @@ multiclass avx512_scalar_fpclass<bits<8> opc, string OpcodeStr, SDNode OpNode,
                                       (OpNode (_.VT _.RC:$src1),
                                       (i32 imm:$src2))))], NoItinerary>, EVEX_K;
     def rm : AVX512<opc, MRMSrcMem, (outs _.KRC:$dst),
-                    (ins _.ScalarMemOp:$src1, i32u8imm:$src2),
+                    (ins _.IntScalarMemOp:$src1, i32u8imm:$src2),
                     OpcodeStr##_.Suffix##
                               "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
                     [(set _.KRC:$dst,
-                          (OpNode (_.VT (scalar_to_vector (_.ScalarLdFrag addr:$src1))),
+                          (OpNode _.ScalarIntMemCPat:$src1,
                                   (i32 imm:$src2)))], NoItinerary>;
     def rmk : AVX512<opc, MRMSrcMem, (outs _.KRC:$dst),
-                    (ins _.KRCWM:$mask, _.ScalarMemOp:$src1, i32u8imm:$src2),
+                    (ins _.KRCWM:$mask, _.IntScalarMemOp:$src1, i32u8imm:$src2),
                     OpcodeStr##_.Suffix##
                     "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}",
                     [(set _.KRC:$dst,(or _.KRCWM:$mask,
-                        (OpNode (_.VT (scalar_to_vector (_.ScalarLdFrag addr:$src1))),
+                        (OpNode _.ScalarIntMemCPat:$src1,
                             (i32 imm:$src2))))], NoItinerary>, EVEX_K;
   }
 }

diff --git a/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll b/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
index 49f439f..529f58d 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512dq-intrinsics.ll
@@ -399,8 +399,7 @@ define i8 @test_int_x86_avx512_mask_fpclass_sd(<2 x double> %x0, i8 %x1) {
 define i8 @test_int_x86_avx512_mask_fpclass_sd_load(<2 x double>* %x0ptr) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_fpclass_sd_load:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vmovapd (%rdi), %xmm0
-; CHECK-NEXT:    vfpclasssd $4, %xmm0, %k0
+; CHECK-NEXT:    vfpclasssd $4, (%rdi), %k0
 ; CHECK-NEXT:    kmovw %k0, %eax
 ; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; CHECK-NEXT:    retq
@@ -431,8 +430,7 @@ define i8 @test_int_x86_avx512_mask_fpclass_ss(<4 x float> %x0, i8 %x1) {
 define i8 @test_int_x86_avx512_mask_fpclass_ss_load(<4 x float>* %x0ptr, i8 %x1) {
 ; CHECK-LABEL: test_int_x86_avx512_mask_fpclass_ss_load:
 ; CHECK:       ## BB#0:
-; CHECK-NEXT:    vmovaps (%rdi), %xmm0
-; CHECK-NEXT:    vfpclassss $4, %xmm0, %k0
+; CHECK-NEXT:    vfpclassss $4, (%rdi), %k0
 ; CHECK-NEXT:    kmovw %k0, %eax
 ; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; CHECK-NEXT:    retq