DAG: Fall back to separate sin and cos when softening sincos

arsenm · arsenm · commit 65011f9bf50b · 2025-07-08T16:24:04.000+09:00
Fix asserting in the error case.
diff --git a/llvm/include/llvm/CodeGen/RuntimeLibcallUtil.h b/llvm/include/llvm/CodeGen/RuntimeLibcallUtil.h
@@ -68,6 +68,14 @@ LLVM_ABI Libcall getLDEXP(EVT RetVT);
 /// UNKNOWN_LIBCALL if there is none.
 LLVM_ABI Libcall getFREXP(EVT RetVT);
 
+/// Return the SIN_* value for the given types, or UNKNOWN_LIBCALL if there is
+/// none.
+LLVM_ABI Libcall getSIN(EVT RetVT);
+
+/// Return the COS_* value for the given types, or UNKNOWN_LIBCALL if there is
+/// none.
+LLVM_ABI Libcall getCOS(EVT RetVT);
+
 /// getSINCOS - Return the SINCOS_* value for the given types, or
 /// UNKNOWN_LIBCALL if there is none.
 LLVM_ABI Libcall getSINCOS(EVT RetVT);
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -794,7 +794,7 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_FFREXP(SDNode *N) {
   return ReturnVal;
 }
 
-SDValue DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(
+bool DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(
     SDNode *N, RTLIB::Libcall LC, std::optional<unsigned> CallRetResNo) {
   assert(!N->isStrictFPOpcode() && "strictfp not implemented");
   EVT VT = N->getValueType(0);
@@ -803,7 +803,7 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(
          "expected both return values to have the same type");
 
   if (!TLI.getLibcallName(LC))
-    return SDValue();
+    return false;
 
   EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
 
@@ -849,17 +849,46 @@ SDValue DAGTypeLegalizer::SoftenFloatRes_UnaryWithTwoFPResults(
     SetSoftenedFloat(SDValue(N, ResNum), CreateStackLoad(SlackSlot));
   }
 
-  return SDValue();
+  return true;
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_FSINCOS(SDNode *N) {
-  return SoftenFloatRes_UnaryWithTwoFPResults(
-      N, RTLIB::getSINCOS(N->getValueType(0)));
+  EVT VT = N->getValueType(0);
+  if (SoftenFloatRes_UnaryWithTwoFPResults(N, RTLIB::getSINCOS(VT)))
+    return SDValue();
+
+  // Fall back on softening the separate sin and cos calls if available.
+  RTLIB::Libcall SinLC = RTLIB::getSIN(VT);
+  RTLIB::Libcall CosLC = RTLIB::getCOS(VT);
+
+  SDValue SoftSin, SoftCos;
+  if (!TLI.getLibcallName(SinLC) || !TLI.getLibcallName(CosLC)) {
+    DAG.getContext()->emitError("do not know how to soften fsincos");
+
+    EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
+    SoftSin = SoftCos = DAG.getPOISON(NVT);
+  } else {
+    SoftSin = SoftenFloatRes_Unary(N, SinLC);
+    SoftCos = SoftenFloatRes_Unary(N, CosLC);
+  }
+
+  SetSoftenedFloat(SDValue(N, 0), SoftSin);
+  SetSoftenedFloat(SDValue(N, 1), SoftCos);
+  return SDValue();
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_FMODF(SDNode *N) {
-  return SoftenFloatRes_UnaryWithTwoFPResults(
-      N, RTLIB::getMODF(N->getValueType(0)), /*CallRetResNo=*/0);
+  EVT VT = N->getValueType(0);
+  if (SoftenFloatRes_UnaryWithTwoFPResults(N, RTLIB::getMODF(VT),
+                                           /*CallRetResNo=*/0))
+    return SDValue();
+
+  EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
+  DAG.getContext()->emitError("do not know how to soften fmodf");
+  SDValue Poison = DAG.getPOISON(NVT);
+  SetSoftenedFloat(SDValue(N, 0), Poison);
+  SetSoftenedFloat(SDValue(N, 1), Poison);
+  return SDValue();
 }
 
 SDValue DAGTypeLegalizer::SoftenFloatRes_FREM(SDNode *N) {
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -564,7 +564,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   // Convert Float Results to Integer.
   void SoftenFloatResult(SDNode *N, unsigned ResNo);
   SDValue SoftenFloatRes_Unary(SDNode *N, RTLIB::Libcall LC);
-  SDValue SoftenFloatRes_UnaryWithTwoFPResults(
+  bool SoftenFloatRes_UnaryWithTwoFPResults(
       SDNode *N, RTLIB::Libcall LC, std::optional<unsigned> CallRetResNo = {});
   SDValue SoftenFloatRes_Binary(SDNode *N, RTLIB::Libcall LC);
   SDValue SoftenFloatRes_MERGE_VALUES(SDNode *N, unsigned ResNo);
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -410,6 +410,14 @@ RTLIB::Libcall RTLIB::getFREXP(EVT RetVT) {
                       FREXP_PPCF128);
 }
 
+RTLIB::Libcall RTLIB::getSIN(EVT RetVT) {
+  return getFPLibCall(RetVT, SIN_F32, SIN_F64, SIN_F80, SIN_F128, SIN_PPCF128);
+}
+
+RTLIB::Libcall RTLIB::getCOS(EVT RetVT) {
+  return getFPLibCall(RetVT, COS_F32, COS_F64, COS_F80, COS_F128, COS_PPCF128);
+}
+
 RTLIB::Libcall RTLIB::getSINCOS(EVT RetVT) {
   return getFPLibCall(RetVT, SINCOS_F32, SINCOS_F64, SINCOS_F80, SINCOS_F128,
                       SINCOS_PPCF128);
diff --git a/llvm/test/CodeGen/AVR/sincos-soften-error.ll b/llvm/test/CodeGen/AVR/sincos-soften-error.ll
@@ -0,0 +1,9 @@
+; RUN: not llc -mtriple=avr -filetype=null %s 2>&1 | FileCheck %s
+
+; CHECK: error: do not know how to soften fsincos
+define { double, double } @test_sincos_f64(double %a) #0 {
+  %result = call { double, double } @llvm.sincos.f64(double %a)
+  ret { double, double } %result
+}
+
+attributes #0 = { nounwind }
diff --git a/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll b/llvm/test/CodeGen/WebAssembly/llvm.sincos.ll
@@ -485,16 +485,232 @@ define { <2 x double>, <2 x double> } @test_sincos_v2f64(<2 x double> %a) #0 {
   ret { <2 x double>, <2 x double> } %result
 }
 
-; ; FIXME: Asserts
-; define { fp128, fp128 } @test_sincos_f128(fp128 %a) #0 {
-;   %result = call { fp128, fp128 } @llvm.sincos.f128(fp128 %a)
-;   ret { fp128, fp128 } %result
-; }
+define { fp128, fp128 } @test_sincos_f128(fp128 %a) #0 {
+; WASM32-LABEL: test_sincos_f128:
+; WASM32:         .functype test_sincos_f128 (i32, i64, i64) -> ()
+; WASM32-NEXT:    .local i32
+; WASM32-NEXT:  # %bb.0:
+; WASM32-NEXT:    global.get __stack_pointer
+; WASM32-NEXT:    i32.const 32
+; WASM32-NEXT:    i32.sub
+; WASM32-NEXT:    local.tee 3
+; WASM32-NEXT:    global.set __stack_pointer
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 2
+; WASM32-NEXT:    call cosl
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i32.const 16
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 2
+; WASM32-NEXT:    call sinl
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i64.load 8
+; WASM32-NEXT:    i64.store 24
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i64.load 0
+; WASM32-NEXT:    i64.store 16
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i64.load 24
+; WASM32-NEXT:    i64.store 8
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i64.load 16
+; WASM32-NEXT:    i64.store 0
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    i32.const 32
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    global.set __stack_pointer
+; WASM32-NEXT:    # fallthrough-return
+;
+; WASM64-LABEL: test_sincos_f128:
+; WASM64:         .functype test_sincos_f128 (i64, i64, i64) -> ()
+; WASM64-NEXT:    .local i64
+; WASM64-NEXT:  # %bb.0:
+; WASM64-NEXT:    global.get __stack_pointer
+; WASM64-NEXT:    i64.const 32
+; WASM64-NEXT:    i64.sub
+; WASM64-NEXT:    local.tee 3
+; WASM64-NEXT:    global.set __stack_pointer
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 2
+; WASM64-NEXT:    call cosl
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.const 16
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 2
+; WASM64-NEXT:    call sinl
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.load 8
+; WASM64-NEXT:    i64.store 24
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.load 0
+; WASM64-NEXT:    i64.store 16
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.load 24
+; WASM64-NEXT:    i64.store 8
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.load 16
+; WASM64-NEXT:    i64.store 0
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    i64.const 32
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    global.set __stack_pointer
+; WASM64-NEXT:    # fallthrough-return
+  %result = call { fp128, fp128 } @llvm.sincos.f128(fp128 %a)
+  ret { fp128, fp128 } %result
+}
 
-; ; FIXME: Asserts
-; define { <2 x fp128>, <2 x fp128> } @test_sincos_v2f128(<2 x fp128> %a) #0 {
-;   %result = call { <2 x fp128>, <2 x fp128> } @llvm.sincos.v2f128(<2 x fp128> %a)
-;   ret { <2 x fp128>, <2 x fp128> } %result
-; }
+define { <2 x fp128>, <2 x fp128> } @test_sincos_v2f128(<2 x fp128> %a) #0 {
+; WASM32-LABEL: test_sincos_v2f128:
+; WASM32:         .functype test_sincos_v2f128 (i32, i64, i64, i64, i64) -> ()
+; WASM32-NEXT:    .local i32
+; WASM32-NEXT:  # %bb.0:
+; WASM32-NEXT:    global.get __stack_pointer
+; WASM32-NEXT:    i32.const 64
+; WASM32-NEXT:    i32.sub
+; WASM32-NEXT:    local.tee 5
+; WASM32-NEXT:    global.set __stack_pointer
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i32.const 32
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    local.get 4
+; WASM32-NEXT:    call cosl
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 2
+; WASM32-NEXT:    call cosl
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i32.const 48
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    local.get 3
+; WASM32-NEXT:    local.get 4
+; WASM32-NEXT:    call sinl
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i32.const 16
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    local.get 1
+; WASM32-NEXT:    local.get 2
+; WASM32-NEXT:    call sinl
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 40
+; WASM32-NEXT:    i64.store 56
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 32
+; WASM32-NEXT:    i64.store 48
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 8
+; WASM32-NEXT:    i64.store 40
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 0
+; WASM32-NEXT:    i64.store 32
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 56
+; WASM32-NEXT:    i64.store 24
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 48
+; WASM32-NEXT:    i64.store 16
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 24
+; WASM32-NEXT:    i64.store 8
+; WASM32-NEXT:    local.get 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i64.load 16
+; WASM32-NEXT:    i64.store 0
+; WASM32-NEXT:    local.get 5
+; WASM32-NEXT:    i32.const 64
+; WASM32-NEXT:    i32.add
+; WASM32-NEXT:    global.set __stack_pointer
+; WASM32-NEXT:    # fallthrough-return
+;
+; WASM64-LABEL: test_sincos_v2f128:
+; WASM64:         .functype test_sincos_v2f128 (i64, i64, i64, i64, i64) -> ()
+; WASM64-NEXT:    .local i64
+; WASM64-NEXT:  # %bb.0:
+; WASM64-NEXT:    global.get __stack_pointer
+; WASM64-NEXT:    i64.const 64
+; WASM64-NEXT:    i64.sub
+; WASM64-NEXT:    local.tee 5
+; WASM64-NEXT:    global.set __stack_pointer
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.const 32
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    local.get 4
+; WASM64-NEXT:    call cosl
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 2
+; WASM64-NEXT:    call cosl
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.const 48
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    local.get 3
+; WASM64-NEXT:    local.get 4
+; WASM64-NEXT:    call sinl
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.const 16
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    local.get 1
+; WASM64-NEXT:    local.get 2
+; WASM64-NEXT:    call sinl
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 40
+; WASM64-NEXT:    i64.store 56
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 32
+; WASM64-NEXT:    i64.store 48
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 8
+; WASM64-NEXT:    i64.store 40
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 0
+; WASM64-NEXT:    i64.store 32
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 56
+; WASM64-NEXT:    i64.store 24
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 48
+; WASM64-NEXT:    i64.store 16
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 24
+; WASM64-NEXT:    i64.store 8
+; WASM64-NEXT:    local.get 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.load 16
+; WASM64-NEXT:    i64.store 0
+; WASM64-NEXT:    local.get 5
+; WASM64-NEXT:    i64.const 64
+; WASM64-NEXT:    i64.add
+; WASM64-NEXT:    global.set __stack_pointer
+; WASM64-NEXT:    # fallthrough-return
+  %result = call { <2 x fp128>, <2 x fp128> } @llvm.sincos.v2f128(<2 x fp128> %a)
+  ret { <2 x fp128>, <2 x fp128> } %result
+}
 
 attributes #0 = { nounwind }
