[SPARC] Legalize truncation and extension between fp128 and half

[LemonBoy]

Summary

Lower truncations and expansions between fp128 and half values into libcalls.
Expand truncating stores into two separate truncation and a store operations.

[jrtc27]

The diff looks fine but:

(a) the commit title only mentions truncstore whilst this also fixes both general truncation and extension
(b) the commit body doesn't mention extension

[LemonBoy]

In D104185#2815375, @jrtc27 wrote:

The diff looks fine but:

(a) the commit title only mentions truncstore whilst this also fixes both general truncation and extension
(b) the commit body doesn't mention extension

I'm bad at picking titles,

In D104185#2815375, @jrtc27 wrote:

The diff looks fine but:

(a) the commit title only mentions truncstore whilst this also fixes both general truncation and extension
(b) the commit body doesn't mention extension

Fixing the extension was a side-effect, I've now updated the commit title and body.

llvm/lib/Target/Sparc/SparcISelLowering.cpp

SparcTargetLowering::SparcTargetLowering(const TargetMachine &TM,
   // Sparc doesn't have i1 sign extending load
   for (MVT VT : MVT::integer_valuetypes())
     setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
   // Turn FP truncstore into trunc + store.
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+  setTruncStoreAction(MVT::f128, MVT::f16, Expand);
   setTruncStoreAction(MVT::f128, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f64, Expand);
 
   // Custom legalize GlobalAddress nodes into LO/HI parts.
   setOperationAction(ISD::GlobalAddress, PtrVT, Custom);
   setOperationAction(ISD::GlobalTLSAddress, PtrVT, Custom);
   setOperationAction(ISD::ConstantPool, PtrVT, Custom);
   setOperationAction(ISD::BlockAddress, PtrVT, Custom);
SparcTargetLowering::SparcTargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
   setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
 
   // Lower f16 conversion operations into library calls
   setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
   setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
   setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
   setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
+  setOperationAction(ISD::FP16_TO_FP, MVT::f128, Expand);
+  setOperationAction(ISD::FP_TO_FP16, MVT::f128, Expand);
 
   setOperationAction(ISD::BITCAST, MVT::f32, Expand);
   setOperationAction(ISD::BITCAST, MVT::i32, Expand);
 
   // Sparc has no select or setcc: expand to SELECT_CC.
   setOperationAction(ISD::SELECT, MVT::i32, Expand);
   setOperationAction(ISD::SELECT, MVT::f32, Expand);
   setOperationAction(ISD::SELECT, MVT::f64, Expand);

llvm/test/CodeGen/SPARC/fp16-promote.ll

 ; SPARC64-NEXT:    fstod %f0, %f0
 ; SPARC64-NEXT:    ret
 ; SPARC64-NEXT:    restore
   %a = load half, half* %p
   %r = fpext half %a to double
   ret double %r
 }
 
+define void @test_fpextend_fp128(half* %p, fp128* %out) nounwind {
+; V8-OPT-LABEL: test_fpextend_fp128:
+; V8-OPT:       ! %bb.0:
+; V8-OPT-NEXT:    save %sp, -112, %sp
+; V8-OPT-NEXT:    call __gnu_h2f_ieee
+; V8-OPT-NEXT:    lduh [%i0], %o0
+; V8-OPT-NEXT:    st %f0, [%fp+-20]
+; V8-OPT-NEXT:    add %fp, -16, %i0
+; V8-OPT-NEXT:    st %i0, [%sp+64]
+; V8-OPT-NEXT:    call _Q_stoq
+; V8-OPT-NEXT:    ld [%fp+-20], %o0
+; V8-OPT-NEXT:    unimp 16
+; V8-OPT-NEXT:    ldd [%fp+-16], %f0
+; V8-OPT-NEXT:    ldd [%fp+-8], %f2
+; V8-OPT-NEXT:    std %f2, [%i1+8]
+; V8-OPT-NEXT:    std %f0, [%i1]
+; V8-OPT-NEXT:    ret
+; V8-OPT-NEXT:    restore
+;
+; V8-UNOPT-LABEL: test_fpextend_fp128:
+; V8-UNOPT:       ! %bb.0:
+; V8-UNOPT-NEXT:    save %sp, -112, %sp
+; V8-UNOPT-NEXT:    call __gnu_h2f_ieee
+; V8-UNOPT-NEXT:    lduh [%i0], %o0
+; V8-UNOPT-NEXT:    st %f0, [%fp+-20]
+; V8-UNOPT-NEXT:    add %fp, -16, %i0
+; V8-UNOPT-NEXT:    st %i0, [%sp+64]
+; V8-UNOPT-NEXT:    call _Q_stoq
+; V8-UNOPT-NEXT:    ld [%fp+-20], %o0
+; V8-UNOPT-NEXT:    unimp 16
+; V8-UNOPT-NEXT:    ldd [%fp+-16], %f4
+; V8-UNOPT-NEXT:    ! implicit-def: $q0
+; V8-UNOPT-NEXT:    fmovs %f4, %f0
+; V8-UNOPT-NEXT:    fmovs %f5, %f1
+; V8-UNOPT-NEXT:    ldd [%fp+-8], %f4
+; V8-UNOPT-NEXT:    fmovs %f4, %f2
+; V8-UNOPT-NEXT:    fmovs %f5, %f3
+; V8-UNOPT-NEXT:    fmovs %f2, %f4
+; V8-UNOPT-NEXT:    fmovs %f3, %f5
+; V8-UNOPT-NEXT:    std %f4, [%i1+8]
+; V8-UNOPT-NEXT:    ! kill: def $d0 killed $d0 killed $q0
+; V8-UNOPT-NEXT:    std %f0, [%i1]
+; V8-UNOPT-NEXT:    ret
+; V8-UNOPT-NEXT:    restore
+;
+; V9-LABEL: test_fpextend_fp128:
+; V9:       ! %bb.0:
+; V9-NEXT:    save %sp, -112, %sp
+; V9-NEXT:    call __gnu_h2f_ieee
+; V9-NEXT:    lduh [%i0], %o0
+; V9-NEXT:    st %f0, [%fp+-20]
+; V9-NEXT:    add %fp, -16, %i0
+; V9-NEXT:    st %i0, [%sp+64]
+; V9-NEXT:    call _Q_stoq
+; V9-NEXT:    ld [%fp+-20], %o0
+; V9-NEXT:    unimp 16
+; V9-NEXT:    ldd [%fp+-16], %f0
+; V9-NEXT:    ldd [%fp+-8], %f2
+; V9-NEXT:    std %f2, [%i1+8]
+; V9-NEXT:    std %f0, [%i1]
+; V9-NEXT:    ret
+; V9-NEXT:    restore
+;
+; SPARC64-LABEL: test_fpextend_fp128:
+; SPARC64:       ! %bb.0:
+; SPARC64-NEXT:    save %sp, -192, %sp
+; SPARC64-NEXT:    call __gnu_h2f_ieee
+; SPARC64-NEXT:    lduh [%i0], %o0
+; SPARC64-NEXT:    add %fp, 2031, %o0
+; SPARC64-NEXT:    fmovs %f0, %f3
+; SPARC64-NEXT:    call _Qp_stoq
+; SPARC64-NEXT:    nop
+; SPARC64-NEXT:    ldd [%fp+2031], %f0
+; SPARC64-NEXT:    ldd [%fp+2039], %f2
+; SPARC64-NEXT:    std %f2, [%i1+8]
+; SPARC64-NEXT:    std %f0, [%i1]
+; SPARC64-NEXT:    ret
+; SPARC64-NEXT:    restore
+  %a = load half, half* %p
+  %r = fpext half %a to fp128
+  store fp128 %r, fp128* %out
+  ret void
+}
+
 define void @test_fptrunc_float(float %f, half* %p) nounwind {
 ; V8-OPT-LABEL: test_fptrunc_float:
 ; V8-OPT:       ! %bb.0:
 ; V8-OPT-NEXT:    save %sp, -96, %sp
 ; V8-OPT-NEXT:    call __gnu_f2h_ieee
 ; V8-OPT-NEXT:    mov %i0, %o0
 ; V8-OPT-NEXT:    sth %o0, [%i1]
 ; V8-OPT-NEXT:    ret

 ; SPARC64-NEXT:    sth %o0, [%i1]
 ; SPARC64-NEXT:    ret
 ; SPARC64-NEXT:    restore
   %a = fptrunc double %d to half
   store half %a, half* %p
   ret void
 }
 
+define void @test_fptrunc_fp128(fp128* %dp, half* %p) nounwind {
+; V8-OPT-LABEL: test_fptrunc_fp128:
+; V8-OPT:       ! %bb.0:
+; V8-OPT-NEXT:    save %sp, -104, %sp
+; V8-OPT-NEXT:    ldd [%i0], %f0
+; V8-OPT-NEXT:    ldd [%i0+8], %f2
+; V8-OPT-NEXT:    std %f2, [%sp+100]
+; V8-OPT-NEXT:    call __trunctfhf2
+; V8-OPT-NEXT:    std %f0, [%sp+92]
+; V8-OPT-NEXT:    sth %o0, [%i1]
+; V8-OPT-NEXT:    ret
+; V8-OPT-NEXT:    restore
+;
+; V8-UNOPT-LABEL: test_fptrunc_fp128:
+; V8-UNOPT:       ! %bb.0:
+; V8-UNOPT-NEXT:    save %sp, -104, %sp
+; V8-UNOPT-NEXT:    ldd [%i0], %f4
+; V8-UNOPT-NEXT:    ! implicit-def: $q0
+; V8-UNOPT-NEXT:    fmovs %f4, %f0
+; V8-UNOPT-NEXT:    fmovs %f5, %f1
+; V8-UNOPT-NEXT:    ldd [%i0+8], %f4
+; V8-UNOPT-NEXT:    fmovs %f4, %f2
+; V8-UNOPT-NEXT:    fmovs %f5, %f3
+; V8-UNOPT-NEXT:    fmovs %f2, %f4
+; V8-UNOPT-NEXT:    fmovs %f3, %f5
+; V8-UNOPT-NEXT:    std %f4, [%sp+100]
+; V8-UNOPT-NEXT:    ! kill: def $d0 killed $d0 killed $q0
+; V8-UNOPT-NEXT:    call __trunctfhf2
+; V8-UNOPT-NEXT:    std %f0, [%sp+92]
+; V8-UNOPT-NEXT:    sth %o0, [%i1]
+; V8-UNOPT-NEXT:    ret
+; V8-UNOPT-NEXT:    restore
+;
+; V9-LABEL: test_fptrunc_fp128:
+; V9:       ! %bb.0:
+; V9-NEXT:    save %sp, -104, %sp
+; V9-NEXT:    ldd [%i0], %f0
+; V9-NEXT:    ldd [%i0+8], %f2
+; V9-NEXT:    std %f2, [%sp+100]
+; V9-NEXT:    call __trunctfhf2
+; V9-NEXT:    std %f0, [%sp+92]
+; V9-NEXT:    sth %o0, [%i1]
+; V9-NEXT:    ret
+; V9-NEXT:    restore
+;
+; SPARC64-LABEL: test_fptrunc_fp128:
+; SPARC64:       ! %bb.0:
+; SPARC64-NEXT:    save %sp, -176, %sp
+; SPARC64-NEXT:    ldd [%i0], %f0
+; SPARC64-NEXT:    call __trunctfhf2
+; SPARC64-NEXT:    ldd [%i0+8], %f2
+; SPARC64-NEXT:    sth %o0, [%i1]
+; SPARC64-NEXT:    ret
+; SPARC64-NEXT:    restore
+  %d = load fp128, fp128* %dp
+  %a = fptrunc fp128 %d to half
+  store half %a, half* %p
+  ret void
+}
+
 define void @test_fadd(half* %p, half* %q) nounwind {
 ; V8-OPT-LABEL: test_fadd:
 ; V8-OPT:       ! %bb.0:
 ; V8-OPT-NEXT:    save %sp, -104, %sp
 ; V8-OPT-NEXT:    call __gnu_h2f_ieee
 ; V8-OPT-NEXT:    lduh [%i0], %o0
 ; V8-OPT-NEXT:    st %f0, [%fp+-8] ! 4-byte Folded Spill
 ; V8-OPT-NEXT:    call __gnu_h2f_ieee