[llvm] [SelectionDAG] Add `STRICT_BF16_TO_FP` and `STRICT_FP_TO_BF16` (PR #80056)

[Phoebe Wang via llvm-commits]

================
@@ -0,0 +1,98 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-linux-gnu | FileCheck %s --check-prefix=X32
+; RUN: llc < %s -mtriple=x86_64-linux-gnu | FileCheck %s --check-prefix=X64
+
+ at a = global bfloat 0xR0000, align 2
+ at b = global bfloat 0xR0000, align 2
+ at c = global bfloat 0xR0000, align 2
+
+; FIXME: We don't have strict extend yet.
+; define float @bfloat_to_float() strictfp {
+;   %1 = load bfloat, ptr @a, align 2
+;   %2 = tail call float @llvm.experimental.constrained.fpext.f32.bfloat(bfloat %1, metadata !"fpexcept.strict") #0
+;   ret float %2
+; }
+
+; define double @bfloat_to_double() strictfp {
+;   %1 = load bfloat, ptr @a, align 2
+;   %2 = tail call double @llvm.experimental.constrained.fpext.f64.bfloat(bfloat %1, metadata !"fpexcept.strict") #0
+;   ret double %2
+; }
+
+define void @float_to_bfloat(float %0) strictfp {
+; X32-LABEL: float_to_bfloat:
+; X32:       # %bb.0:
+; X32-NEXT:    subl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 16
+; X32-NEXT:    flds {{[0-9]+}}(%esp)
+; X32-NEXT:    fstps (%esp)
+; X32-NEXT:    wait
+; X32-NEXT:    calll __truncsfbf2
+; X32-NEXT:    movw %ax, a
+; X32-NEXT:    addl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 4
+; X32-NEXT:    retl
+;
+; X64-LABEL: float_to_bfloat:
+; X64:       # %bb.0:
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    callq __truncsfbf2 at PLT
+; X64-NEXT:    movq a at GOTPCREL(%rip), %rcx
+; X64-NEXT:    movw %ax, (%rcx)
+; X64-NEXT:    popq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 8
+; X64-NEXT:    retq
+  %2 = tail call bfloat @llvm.experimental.constrained.fptrunc.bfloat.f32(float %0, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
+  store bfloat %2, ptr @a, align 2
+  ret void
+}
+
+define void @double_to_bfloat(double %0) strictfp {
+; X32-LABEL: double_to_bfloat:
+; X32:       # %bb.0:
+; X32-NEXT:    subl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 16
+; X32-NEXT:    fldl {{[0-9]+}}(%esp)
+; X32-NEXT:    fstpl (%esp)
+; X32-NEXT:    wait
+; X32-NEXT:    calll __truncdfbf2
+; X32-NEXT:    movw %ax, a
+; X32-NEXT:    addl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 4
+; X32-NEXT:    retl
+;
+; X64-LABEL: double_to_bfloat:
+; X64:       # %bb.0:
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 16
+; X64-NEXT:    callq __truncdfbf2 at PLT
+; X64-NEXT:    movq a at GOTPCREL(%rip), %rcx
+; X64-NEXT:    movw %ax, (%rcx)
+; X64-NEXT:    popq %rax
+; X64-NEXT:    .cfi_def_cfa_offset 8
+; X64-NEXT:    retq
+  %2 = tail call bfloat @llvm.experimental.constrained.fptrunc.bfloat.f64(double %0, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
+  store bfloat %2, ptr @a, align 2
+  ret void
+}
+
+; define void @add() strictfp {
+;   %1 = load bfloat, ptr @a, align 2
+;   %2 = tail call float @llvm.experimental.constrained.fpext.f32.bfloat(bfloat %1, metadata !"fpexcept.strict") #0
+;   %3 = load bfloat, ptr @b, align 2
+;   %4 = tail call float @llvm.experimental.constrained.fpext.f32.bfloat(bfloat %3, metadata !"fpexcept.strict") #0
+;   %5 = tail call float @llvm.experimental.constrained.fadd.f32(float %2, float %4, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
+;   %6 = tail call bfloat @llvm.experimental.constrained.fptrunc.bfloat.f32(float %5, metadata !"round.tonearest", metadata !"fpexcept.strict") #0
+;   store bfloat %6, ptr @c, align 2
+;   ret void
+; }
+
+; declare float @llvm.experimental.constrained.fpext.f32.bfloat(bfloat, metadata)
+; declare double @llvm.experimental.constrained.fpext.f64.bfloat(bfloat, metadata)
+; declare float @llvm.experimental.constrained.fadd.f32(float, float, metadata, metadata)
+declare bfloat @llvm.experimental.constrained.fptrunc.bfloat.f32(float, metadata, metadata)
+declare bfloat @llvm.experimental.constrained.fptrunc.bfloat.f64(double, metadata, metadata)
+
----------------
phoebewang wrote:

Yes, we may need to add it to compiler-rt too.

https://github.com/llvm/llvm-project/pull/80056