summaryrefslogtreecommitdiffstats
path: root/llvm/lib
diff options
context:
space:
mode:
authorMatt Arsenault <Matthew.Arsenault@amd.com>2016-05-28 00:19:52 +0000
committerMatt Arsenault <Matthew.Arsenault@amd.com>2016-05-28 00:19:52 +0000
commit7401516985bcffeb67db966028d4279058c5b05d (patch)
tree42dbf628e6c88717ee6c1ab1d69e0b346047cb7b /llvm/lib
parent5a7f09c5af24eca92cd2acf701734009351a0fd3 (diff)
downloadbcm5719-llvm-7401516985bcffeb67db966028d4279058c5b05d.tar.gz
bcm5719-llvm-7401516985bcffeb67db966028d4279058c5b05d.zip
AMDGPU: Add fract intrinsic
Remove broken patterns matching it. This was matching the unsafe math pattern and expanding the fix for the buggy instruction from the pattern. The problems are also on CI. Remove the workarounds and only use fract with unsafe math or from the intrinsic. llvm-svn: 271078
Diffstat (limited to 'llvm/lib')
-rw-r--r--llvm/lib/Target/AMDGPU/CIInstructions.td19
-rw-r--r--llvm/lib/Target/AMDGPU/SIISelLowering.cpp4
-rw-r--r--llvm/lib/Target/AMDGPU/SIInstructions.td34
3 files changed, 21 insertions, 36 deletions
diff --git a/llvm/lib/Target/AMDGPU/CIInstructions.td b/llvm/lib/Target/AMDGPU/CIInstructions.td
index 6adc01372f4..9fb6e6039e5 100644
--- a/llvm/lib/Target/AMDGPU/CIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/CIInstructions.td
@@ -258,25 +258,6 @@ defm FLAT_ATOMIC_FMAX_X2 : FLAT_ATOMIC <
} // End SubtargetPredicate = isCI, VIAssemblerPredicate = DisableInst, DisableVIDecoder = 1
-let Predicates = [isCI] in {
-
-// Convert (x - floor(x)) to fract(x)
-def : Pat <
- (f32 (fsub (f32 (VOP3Mods f32:$x, i32:$mods)),
- (f32 (ffloor (f32 (VOP3Mods f32:$x, i32:$mods)))))),
- (V_FRACT_F32_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
->;
-
-// Convert (x + (-floor(x))) to fract(x)
-def : Pat <
- (f64 (fadd (f64 (VOP3Mods f64:$x, i32:$mods)),
- (f64 (fneg (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))))))),
- (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
->;
-
-} // End Predicates = [isCI]
-
-
//===----------------------------------------------------------------------===//
// Flat Patterns
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index c4e64dcda47..bc5fc6590f1 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1656,6 +1656,10 @@ SDValue SITargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
case Intrinsic::amdgcn_ldexp:
return DAG.getNode(AMDGPUISD::LDEXP, DL, VT,
Op.getOperand(1), Op.getOperand(2));
+
+ case Intrinsic::amdgcn_fract:
+ return DAG.getNode(AMDGPUISD::FRACT, DL, VT, Op.getOperand(1));
+
case Intrinsic::amdgcn_class:
return DAG.getNode(AMDGPUISD::FP_CLASS, DL, VT,
Op.getOperand(1), Op.getOperand(2));
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index b8e68715a41..4403c262702 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -1367,7 +1367,7 @@ defm V_FREXP_MANT_F64 : VOP1Inst <vop1<0x3d, 0x31>, "v_frexp_mant_f64",
>;
defm V_FRACT_F64 : VOP1Inst <vop1<0x3e, 0x32>, "v_fract_f64",
- VOP_F64_F64
+ VOP_F64_F64, AMDGPUfract
>;
} // End SchedRW = [WriteDoubleAdd]
@@ -2469,7 +2469,22 @@ let Predicates = [UnsafeFPMath] in {
def : RsqPat<V_RSQ_F32_e32, f32>;
def : RsqPat<V_RSQ_F64_e32, f64>;
-}
+
+// Convert (x - floor(x)) to fract(x)
+def : Pat <
+ (f32 (fsub (f32 (VOP3Mods f32:$x, i32:$mods)),
+ (f32 (ffloor (f32 (VOP3Mods f32:$x, i32:$mods)))))),
+ (V_FRACT_F32_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
+>;
+
+// Convert (x + (-floor(x))) to fract(x)
+def : Pat <
+ (f64 (fadd (f64 (VOP3Mods f64:$x, i32:$mods)),
+ (f64 (fneg (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))))))),
+ (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE)
+>;
+
+} // End Predicates = [UnsafeFPMath]
//===----------------------------------------------------------------------===//
// VOP2 Patterns
@@ -3549,21 +3564,6 @@ let Predicates = [isSI] in {
// The workaround for the V_FRACT bug is:
// fract(x) = isnan(x) ? x : min(V_FRACT(x), 0.99999999999999999)
-// Convert (x + (-floor(x)) to fract(x)
-def : Pat <
- (f64 (fadd (f64 (VOP3Mods f64:$x, i32:$mods)),
- (f64 (fneg (f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))))))),
- (V_CNDMASK_B64_PSEUDO
- (V_MIN_F64
- SRCMODS.NONE,
- (V_FRACT_F64_e64 $mods, $x, DSTCLAMP.NONE, DSTOMOD.NONE),
- SRCMODS.NONE,
- (V_MOV_B64_PSEUDO 0x3fefffffffffffff),
- DSTCLAMP.NONE, DSTOMOD.NONE),
- $x,
- (V_CMP_CLASS_F64_e64 SRCMODS.NONE, $x, 3/*NaN*/))
->;
-
// Convert floor(x) to (x - fract(x))
def : Pat <
(f64 (ffloor (f64 (VOP3Mods f64:$x, i32:$mods)))),
OpenPOWER on IntegriCloud