[X86] Use sse_load_f32/f64 and timm in patterns for memory form of vgetmantss/sd.

Previously we only matched scalar_to_vector and scalar load, but
we should be able to narrow a vector load or match vzload.

Also need to match TargetConstant instead of Constant. The register
patterns were previously updated, but not the memory patterns.

llvm-svn: 372458

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index cd833b7..cb0d009 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -10280,12 +10280,11 @@ multiclass avx512_fp_scalar_imm<bits<8> opc, string OpcodeStr, SDNode OpNode,
                               (i32 timm:$src3))>,
                       Sched<[sched]>;
   defm rmi : AVX512_maskable_scalar<opc, MRMSrcMem, _, (outs _.RC:$dst),
-                    (ins _.RC:$src1, _.ScalarMemOp:$src2, i32u8imm:$src3),
+                    (ins _.RC:$src1, _.IntScalarMemOp:$src2, i32u8imm:$src3),
                     OpcodeStr, "$src3, $src2, $src1", "$src1, $src2, $src3",
                     (OpNode (_.VT _.RC:$src1),
-                            (_.VT (scalar_to_vector
-                                      (_.ScalarLdFrag addr:$src2))),
-                            (i32 imm:$src3))>,
+                            (_.VT _.ScalarIntMemCPat:$src2),
+                            (i32 timm:$src3))>,
                     Sched<[sched.Folded, sched.ReadAfterFold]>;
   }
 }

diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
index 90b211f..e94f16a 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics.ll
@@ -4801,15 +4801,13 @@ define <4 x float>@test_int_x86_avx512_mask_getmant_ss(<4 x float> %x0, <4 x flo
 define <4 x float> @test_int_x86_avx512_mask_getmant_ss_load(<4 x float> %x0, <4 x float>* %x1p) {
 ; X64-LABEL: test_int_x86_avx512_mask_getmant_ss_load:
 ; X64:       # %bb.0:
-; X64-NEXT:    vmovaps (%rdi), %xmm1
-; X64-NEXT:    vgetmantss $11, %xmm1, %xmm0, %xmm0
+; X64-NEXT:    vgetmantss $11, (%rdi), %xmm0, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: test_int_x86_avx512_mask_getmant_ss_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vmovaps (%eax), %xmm1
-; X86-NEXT:    vgetmantss $11, %xmm1, %xmm0, %xmm0
+; X86-NEXT:    vgetmantss $11, (%eax), %xmm0, %xmm0
 ; X86-NEXT:    retl
   %x1 = load <4 x float>, <4 x float>* %x1p
   %res = call <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float> %x0, <4 x float> %x1, i32 11, <4 x float> undef, i8 -1, i32 4)