[X86][AVX512F] Fix reg class for VMOVSSZrr/VMOVSSZrrk and VMOVSDZrr/VMOVSDZrrk

[RKSimon]

Summary

Fixed -verify-machineinstrs errors in fast-isel-select-sse.ll (one of many in PR27481)

The VMOVSSZrr/VMOVSSZrrk and VMOVSDZrr/VMOVSDZrrk instructions were assuming both source registers were V128X when the second is actually supposed to be FR32X/FR64X

[craig.topper]

Aren't there now broken patterns using COPY_TO_REGCLASS from FR32X/FR64X to VR128X below this now? For example the avx512_move_scalar_lowering multiclass.

[RKSimon]

Fixed missed patterns

[craig.topper]

LGTM

llvm/trunk/lib/Target/X86/X86InstrAVX512.td

multiclass avx512_move_scalar<string asm, SDNode OpNode,
                               X86VectorVTInfo _> {
   def rr : AVX512PI<0x10, MRMSrcReg, (outs _.RC:$dst),
              (ins _.RC:$src1, _.FRC:$src2),
              !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
              [(set _.RC:$dst, (_.VT (OpNode _.RC:$src1,
                                     (scalar_to_vector _.FRC:$src2))))],
              _.ExeDomain,IIC_SSE_MOV_S_RR>, EVEX_4V;
   def rrkz : AVX512PI<0x10, MRMSrcReg, (outs _.RC:$dst),
-              (ins _.KRCWM:$mask, _.RC:$src1, _.RC:$src2),
+              (ins _.KRCWM:$mask, _.RC:$src1, _.FRC:$src2),
               !strconcat(asm, "\t{$src2, $src1, $dst {${mask}} {z}|",
               "$dst {${mask}} {z}, $src1, $src2}"),
               [(set _.RC:$dst, (_.VT (X86selects _.KRCWM:$mask,
-                                      (_.VT (OpNode _.RC:$src1, _.RC:$src2)),
+                                      (_.VT (OpNode _.RC:$src1,
+                                            (scalar_to_vector _.FRC:$src2))),
                                       _.ImmAllZerosV)))],
               _.ExeDomain,IIC_SSE_MOV_S_RR>, EVEX_4V, EVEX_KZ;
   let Constraints = "$src0 = $dst"  in
   def rrk : AVX512PI<0x10, MRMSrcReg, (outs _.RC:$dst),
-             (ins _.RC:$src0, _.KRCWM:$mask, _.RC:$src1, _.RC:$src2),
+             (ins _.RC:$src0, _.KRCWM:$mask, _.RC:$src1, _.FRC:$src2),
              !strconcat(asm, "\t{$src2, $src1, $dst {${mask}}|",
              "$dst {${mask}}, $src1, $src2}"),
              [(set _.RC:$dst, (_.VT (X86selects _.KRCWM:$mask,
-                                     (_.VT (OpNode _.RC:$src1, _.RC:$src2)),
+                                     (_.VT (OpNode _.RC:$src1,
+                                           (scalar_to_vector _.FRC:$src2))),
                                      (_.VT _.RC:$src0))))],
              _.ExeDomain,IIC_SSE_MOV_S_RR>, EVEX_4V, EVEX_K;
   let canFoldAsLoad = 1, isReMaterializable = 1 in
   def rm : AVX512PI<0x10, MRMSrcMem, (outs _.FRC:$dst), (ins _.ScalarMemOp:$src),
              !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
              [(set _.FRC:$dst, (_.ScalarLdFrag addr:$src))],
              _.ExeDomain, IIC_SSE_MOV_S_RM>, EVEX;
   let mayLoad = 1, hasSideEffects = 0 in {

 def : Pat<(_.VT (OpNode _.RC:$src0,
                         (_.VT (scalar_to_vector
                                   (_.EltVT (X86selects (i1 (trunc GR32:$mask)),
                                                        (_.EltVT _.FRC:$src1),
                                                        (_.EltVT _.FRC:$src2))))))),
           (COPY_TO_REGCLASS (!cast<Instruction>(InstrStr#rrk)
                                           (COPY_TO_REGCLASS _.FRC:$src2, _.RC),
                                           (COPY_TO_REGCLASS GR32:$mask, VK1WM),
-                                          (_.VT _.RC:$src0),
-                                          (COPY_TO_REGCLASS _.FRC:$src1, _.RC)),
+                                          (_.VT _.RC:$src0), _.FRC:$src1),
                             _.RC)>;
 
 def : Pat<(_.VT (OpNode _.RC:$src0,
                         (_.VT (scalar_to_vector
                                   (_.EltVT (X86selects (i1 (trunc GR32:$mask)),
                                                        (_.EltVT _.FRC:$src1),
                                                        (_.EltVT ZeroFP))))))),
           (COPY_TO_REGCLASS (!cast<Instruction>(InstrStr#rrkz)
                                           (COPY_TO_REGCLASS GR32:$mask, VK1WM),
-                                          (_.VT _.RC:$src0),
-                                          (COPY_TO_REGCLASS _.FRC:$src1, _.RC)),
+                                          (_.VT _.RC:$src0), _.FRC:$src1),
                             _.RC)>;
-
 }
 
 multiclass avx512_store_scalar_lowering<string InstrStr, AVX512VLVectorVTInfo _,
                                         dag Mask, RegisterClass MaskRC> {
 
 def : Pat<(masked_store addr:$dst, Mask,
              (_.info512.VT (insert_subvector undef,
                                (_.info256.VT (insert_subvector undef,
defm : avx512_load_scalar_lowering<"VMOVSSZ", avx512vl_f32_info,
                    (v16i1 (bitconvert (i16 (trunc (and GR32:$mask, (i32 1)))))), GR32>;
 defm : avx512_load_scalar_lowering<"VMOVSSZ", avx512vl_f32_info,
                    (v16i1 (bitconvert (i16 (and GR16:$mask, (i16 1))))), GR16>;
 defm : avx512_load_scalar_lowering<"VMOVSDZ", avx512vl_f64_info,
                    (v8i1 (bitconvert (i8 (and GR8:$mask, (i8 1))))), GR8>;
 
 def : Pat<(f32 (X86selects VK1WM:$mask, (f32 FR32X:$src1), (f32 FR32X:$src2))),
           (COPY_TO_REGCLASS (VMOVSSZrrk (COPY_TO_REGCLASS FR32X:$src2, VR128X),
-           VK1WM:$mask, (v4f32 (IMPLICIT_DEF)),(COPY_TO_REGCLASS FR32X:$src1, VR128X)), FR32X)>;
+           VK1WM:$mask, (v4f32 (IMPLICIT_DEF)), FR32X:$src1), FR32X)>;
 
 def : Pat<(f64 (X86selects VK1WM:$mask, (f64 FR64X:$src1), (f64 FR64X:$src2))),
           (COPY_TO_REGCLASS (VMOVSDZrrk (COPY_TO_REGCLASS FR64X:$src2, VR128X),
-           VK1WM:$mask, (v2f64 (IMPLICIT_DEF)), (COPY_TO_REGCLASS FR64X:$src1, VR128X)), FR64X)>;
+           VK1WM:$mask, (v2f64 (IMPLICIT_DEF)), FR64X:$src1), FR64X)>;
 
 def : Pat<(int_x86_avx512_mask_store_ss addr:$dst, VR128X:$src, GR8:$mask),
           (VMOVSSZmrk addr:$dst, (i1 (COPY_TO_REGCLASS GR8:$mask, VK1WM)),
            (COPY_TO_REGCLASS VR128X:$src, FR32X))>;
 
 let hasSideEffects = 0 in
 defm VMOVSSZrr_REV : AVX512_maskable_in_asm<0x11, MRMDestReg, f32x_info,
                            (outs VR128X:$dst), (ins VR128X:$src1, FR32X:$src2),

llvm/trunk/test/CodeGen/X86/fast-isel-select-sse.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown                                          | FileCheck %s --check-prefix=SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -fast-isel -fast-isel-abort=1            | FileCheck %s --check-prefix=SSE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown                               -mattr=avx | FileCheck %s --check-prefix=AVX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -fast-isel -fast-isel-abort=1 -mattr=avx | FileCheck %s --check-prefix=AVX
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown                               -mattr=avx512f | FileCheck %s --check-prefix=AVX512
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -fast-isel -fast-isel-abort=1 -mattr=avx512f | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs                                          | FileCheck %s --check-prefix=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs -fast-isel -fast-isel-abort=1            | FileCheck %s --check-prefix=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs                               -mattr=avx | FileCheck %s --check-prefix=AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs -fast-isel -fast-isel-abort=1 -mattr=avx | FileCheck %s --check-prefix=AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs                               -mattr=avx512f | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs -fast-isel -fast-isel-abort=1 -mattr=avx512f | FileCheck %s --check-prefix=AVX512
 
 ; Test all cmp predicates that can be used with SSE.
 
 define float @select_fcmp_oeq_f32(float %a, float %b, float %c, float %d) {
 ; SSE-LABEL: select_fcmp_oeq_f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    cmpeqss %xmm1, %xmm0
 ; SSE-NEXT:    andps %xmm0, %xmm2