[llvm] [IR] Add `llvm.sincos` intrinsic (PR #109825)

[Benjamin Maxwell via llvm-commits]

================
@@ -0,0 +1,936 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=aarch64-gnu-linux < %s | FileCheck -check-prefixes=CHECK %s
+; RUN: llc -mtriple=aarch64-none-linux < %s | FileCheck -check-prefixes=NO-LIBCALL %s
+
+define { half, half } @test_sincos_f16(half %a) {
+; CHECK-LABEL: test_sincos_f16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset w30, -16
+; CHECK-NEXT:    fcvt s0, h0
+; CHECK-NEXT:    add x0, sp, #12
+; CHECK-NEXT:    add x1, sp, #8
+; CHECK-NEXT:    bl sincosf
+; CHECK-NEXT:    ldp s1, s0, [sp, #8]
+; CHECK-NEXT:    fcvt h0, s0
+; CHECK-NEXT:    fcvt h1, s1
+; CHECK-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
+;
+; NO-LIBCALL-LABEL: test_sincos_f16:
+; NO-LIBCALL:       // %bb.0:
+; NO-LIBCALL-NEXT:    stp d9, d8, [sp, #-32]! // 16-byte Folded Spill
+; NO-LIBCALL-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
+; NO-LIBCALL-NEXT:    .cfi_def_cfa_offset 32
+; NO-LIBCALL-NEXT:    .cfi_offset w30, -16
+; NO-LIBCALL-NEXT:    .cfi_offset b8, -24
+; NO-LIBCALL-NEXT:    .cfi_offset b9, -32
+; NO-LIBCALL-NEXT:    fcvt s8, h0
+; NO-LIBCALL-NEXT:    fmov s0, s8
+; NO-LIBCALL-NEXT:    bl sinf
+; NO-LIBCALL-NEXT:    fcvt h9, s0
+; NO-LIBCALL-NEXT:    fmov s0, s8
+; NO-LIBCALL-NEXT:    bl cosf
+; NO-LIBCALL-NEXT:    fmov s1, s0
+; NO-LIBCALL-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
+; NO-LIBCALL-NEXT:    fmov s0, s9
+; NO-LIBCALL-NEXT:    fcvt h1, s1
+; NO-LIBCALL-NEXT:    ldp d9, d8, [sp], #32 // 16-byte Folded Reload
+; NO-LIBCALL-NEXT:    ret
+  %result = call { half, half } @llvm.sincos.f16(half %a)
+  ret { half, half } %result
+}
+
+define half @test_sincos_f16_only_use_sin(half %a) {
+; CHECK-LABEL: test_sincos_f16_only_use_sin:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset w30, -16
+; CHECK-NEXT:    fcvt s0, h0
+; CHECK-NEXT:    add x0, sp, #12
+; CHECK-NEXT:    add x1, sp, #8
+; CHECK-NEXT:    bl sincosf
+; CHECK-NEXT:    ldr s0, [sp, #12]
+; CHECK-NEXT:    fcvt h0, s0
+; CHECK-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
+;
+; NO-LIBCALL-LABEL: test_sincos_f16_only_use_sin:
+; NO-LIBCALL:       // %bb.0:
+; NO-LIBCALL-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; NO-LIBCALL-NEXT:    .cfi_def_cfa_offset 16
+; NO-LIBCALL-NEXT:    .cfi_offset w30, -16
+; NO-LIBCALL-NEXT:    fcvt s0, h0
+; NO-LIBCALL-NEXT:    bl sinf
+; NO-LIBCALL-NEXT:    fcvt h0, s0
+; NO-LIBCALL-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; NO-LIBCALL-NEXT:    ret
+  %result = call { half, half } @llvm.sincos.f16(half %a)
+  %result.0 = extractvalue { half, half } %result, 0
+  ret half %result.0
+}
+
+define half @test_sincos_f16_only_use_cos(half %a) {
+; CHECK-LABEL: test_sincos_f16_only_use_cos:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset w30, -16
+; CHECK-NEXT:    fcvt s0, h0
+; CHECK-NEXT:    add x0, sp, #12
+; CHECK-NEXT:    add x1, sp, #8
+; CHECK-NEXT:    bl sincosf
+; CHECK-NEXT:    ldr s0, [sp, #8]
+; CHECK-NEXT:    fcvt h0, s0
+; CHECK-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
+;
+; NO-LIBCALL-LABEL: test_sincos_f16_only_use_cos:
+; NO-LIBCALL:       // %bb.0:
+; NO-LIBCALL-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
+; NO-LIBCALL-NEXT:    .cfi_def_cfa_offset 16
+; NO-LIBCALL-NEXT:    .cfi_offset w30, -16
+; NO-LIBCALL-NEXT:    fcvt s0, h0
+; NO-LIBCALL-NEXT:    bl cosf
+; NO-LIBCALL-NEXT:    fcvt h0, s0
+; NO-LIBCALL-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
+; NO-LIBCALL-NEXT:    ret
+  %result = call { half, half } @llvm.sincos.f16(half %a)
+  %result.1 = extractvalue { half, half } %result, 1
+  ret half %result.1
+}
+
----------------
MacDue wrote:

Added this test now :+1: 

https://github.com/llvm/llvm-project/pull/109825