https://github.com/ecnelises updated
https://github.com/llvm/llvm-project/pull/67302
>From a1567f579531c3abbd1f4e9b7c7edd2f95ead42c Mon Sep 17 00:00:00 2001
From: Qiu Chaofan
Date: Mon, 25 Sep 2023 17:10:51 +0800
Subject: [PATCH 1/7] [PowerPC] Implement llvm.set.rounding intrinsic
According to
@@ -8900,6 +8900,82 @@ SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
return FP;
}
+SDValue PPCTargetLowering::LowerSET_ROUNDING(SDValue Op,
+ SelectionDAG &DAG) const {
+ SDLoc Dl(Op);
+ MachineFunction &MF = DAG.getMachi
ecnelises wrote:
> Maybe we can do some perf test between this expansion for set rounding mode
> and the system library's version for fesetround().
They are faster than system `fesetround` on both Linux and AIX. Linux glibc
optimizes `fesetround` with faster `mffscrn` on P9, I just exploited t
https://github.com/ecnelises updated
https://github.com/llvm/llvm-project/pull/67302
>From a1567f579531c3abbd1f4e9b7c7edd2f95ead42c Mon Sep 17 00:00:00 2001
From: Qiu Chaofan
Date: Mon, 25 Sep 2023 17:10:51 +0800
Subject: [PATCH 1/6] [PowerPC] Implement llvm.set.rounding intrinsic
According to
@@ -8900,6 +8900,82 @@ SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
return FP;
}
+SDValue PPCTargetLowering::LowerSET_ROUNDING(SDValue Op,
+ SelectionDAG &DAG) const {
+ SDLoc Dl(Op);
+ MachineFunction &MF = DAG.getMachi
@@ -77,4 +77,196 @@ return: ; preds = %entry
ret i32 %retval3
}
-declare i32 @llvm.get.rounding() nounwind
+define void @setrnd_tozero() {
+; PPC32-LABEL: setrnd_tozero:
+; PPC32: # %bb.0: # %entry
+; PPC32-NEXT:mtfsb0 30
+; PPC32-NEXT:mtfsb1 3
https://github.com/ecnelises updated
https://github.com/llvm/llvm-project/pull/67302
>From a1567f579531c3abbd1f4e9b7c7edd2f95ead42c Mon Sep 17 00:00:00 2001
From: Qiu Chaofan
Date: Mon, 25 Sep 2023 17:10:51 +0800
Subject: [PATCH 1/4] [PowerPC] Implement llvm.set.rounding intrinsic
According to
