diff options
Diffstat (limited to 'lib')
| -rw-r--r-- | lib/Target/PTX/PTXInstrInfo.td | 30 |
1 files changed, 18 insertions, 12 deletions
diff --git a/lib/Target/PTX/PTXInstrInfo.td b/lib/Target/PTX/PTXInstrInfo.td index d497283b72..d240899c6a 100644 --- a/lib/Target/PTX/PTXInstrInfo.td +++ b/lib/Target/PTX/PTXInstrInfo.td @@ -826,31 +826,35 @@ defm STs : PTX_ST_ALL<"st.shared", store_shared>; // TODO: Do something with st.param if/when it is needed. // Conversion to pred - +// PTX does not directly support converting to a predicate type, so we fake it +// by performing a greater-than test between the value and zero. This follows +// the C convention that any non-zero value is equivalent to 'true'. def CVT_pred_u16 - : InstPTX<(outs RegPred:$d), (ins RegI16:$a), "cvt.pred.u16\t$d, $a", + : InstPTX<(outs RegPred:$d), (ins RegI16:$a), "setp.gt.b16\t$d, $a, 0", [(set RegPred:$d, (trunc RegI16:$a))]>; def CVT_pred_u32 - : InstPTX<(outs RegPred:$d), (ins RegI32:$a), "cvt.pred.u32\t$d, $a", + : InstPTX<(outs RegPred:$d), (ins RegI32:$a), "setp.gt.b32\t$d, $a, 0", [(set RegPred:$d, (trunc RegI32:$a))]>; def CVT_pred_u64 - : InstPTX<(outs RegPred:$d), (ins RegI64:$a), "cvt.pred.u64\t$d, $a", + : InstPTX<(outs RegPred:$d), (ins RegI64:$a), "setp.gt.b64\t$d, $a, 0", [(set RegPred:$d, (trunc RegI64:$a))]>; def CVT_pred_f32 - : InstPTX<(outs RegPred:$d), (ins RegF32:$a), "cvt.rzi.pred.f32\t$d, $a", + : InstPTX<(outs RegPred:$d), (ins RegF32:$a), "setp.gt.b32\t$d, $a, 0", [(set RegPred:$d, (fp_to_uint RegF32:$a))]>; def CVT_pred_f64 - : InstPTX<(outs RegPred:$d), (ins RegF64:$a), "cvt.rzi.pred.f64\t$d, $a", + : InstPTX<(outs RegPred:$d), (ins RegF64:$a), "setp.gt.b64\t$d, $a, 0", [(set RegPred:$d, (fp_to_uint RegF64:$a))]>; // Conversion to u16 - +// PTX does not directly support converting a predicate to a value, so we +// use a select instruction to select either 0 or 1 (integer or fp) based +// on the truth value of the predicate. def CVT_u16_pred - : InstPTX<(outs RegI16:$d), (ins RegPred:$a), "cvt.u16.pred\t$d, $a", + : InstPTX<(outs RegI16:$d), (ins RegPred:$a), "selp.u16\t$d, 1, 0, $a", [(set RegI16:$d, (zext RegPred:$a))]>; def CVT_u16_u32 @@ -872,7 +876,7 @@ def CVT_u16_f64 // Conversion to u32 def CVT_u32_pred - : InstPTX<(outs RegI32:$d), (ins RegPred:$a), "cvt.u32.pred\t$d, $a", + : InstPTX<(outs RegI32:$d), (ins RegPred:$a), "selp.u32\t$d, 1, 0, $a", [(set RegI32:$d, (zext RegPred:$a))]>; def CVT_u32_u16 @@ -894,7 +898,7 @@ def CVT_u32_f64 // Conversion to u64 def CVT_u64_pred - : InstPTX<(outs RegI64:$d), (ins RegPred:$a), "cvt.u64.pred\t$d, $a", + : InstPTX<(outs RegI64:$d), (ins RegPred:$a), "selp.u64\t$d, 1, 0, $a", [(set RegI64:$d, (zext RegPred:$a))]>; def CVT_u64_u16 @@ -916,7 +920,8 @@ def CVT_u64_f64 // Conversion to f32 def CVT_f32_pred - : InstPTX<(outs RegF32:$d), (ins RegPred:$a), "cvt.rn.f32.pred\t$d, $a", + : InstPTX<(outs RegF32:$d), (ins RegPred:$a), + "selp.f32\t$d, 0F3F800000, 0F00000000, $a", // 1.0 [(set RegF32:$d, (uint_to_fp RegPred:$a))]>; def CVT_f32_u16 @@ -938,7 +943,8 @@ def CVT_f32_f64 // Conversion to f64 def CVT_f64_pred - : InstPTX<(outs RegF64:$d), (ins RegPred:$a), "cvt.rn.f64.pred\t$d, $a", + : InstPTX<(outs RegF64:$d), (ins RegPred:$a), + "selp.f64\t$d, 0D3F80000000000000, 0D0000000000000000, $a", // 1.0 [(set RegF64:$d, (uint_to_fp RegPred:$a))]>; def CVT_f64_u16 |