[llvm-commits] [llvm] r127578 - in /llvm/trunk: lib/Target/PTX/PTXInstrInfo.td test/CodeGen/PTX/llvm-intrinsic.ll

[Justin Holewinski]

Author: jholewinski
Date: Mon Mar 14 09:09:33 2011
New Revision: 127578

URL: http://llvm.org/viewvc/llvm-project?rev=127578&view=rev
Log:
PTX: Add support for sqrt/sin/cos intrinsics

Added:
    llvm/trunk/test/CodeGen/PTX/llvm-intrinsic.ll
Modified:
    llvm/trunk/lib/Target/PTX/PTXInstrInfo.td

Modified: llvm/trunk/lib/Target/PTX/PTXInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PTX/PTXInstrInfo.td?rev=127578&r1=127577&r2=127578&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PTX/PTXInstrInfo.td (original)
+++ llvm/trunk/lib/Target/PTX/PTXInstrInfo.td Mon Mar 14 09:09:33 2011
@@ -414,6 +414,39 @@
 // TODO: Allow the rounding mode to be selectable through llc.
 defm FMAD : PTX_FLOAT_4OP<"mad.rn", fmul, fadd>;
 
+///===- Floating-Point Intrinsic Instructions -----------------------------===//
+
+def FSQRT32 : InstPTX<(outs RRegf32:$d),
+                      (ins RRegf32:$a),
+                      "sqrt.rn.f32\t$d, $a",
+                      [(set RRegf32:$d, (fsqrt RRegf32:$a))]>;
+
+def FSQRT64 : InstPTX<(outs RRegf64:$d),
+                      (ins RRegf64:$a),
+                      "sqrt.rn.f64\t$d, $a",
+                      [(set RRegf64:$d, (fsqrt RRegf64:$a))]>;
+
+def FSIN32 : InstPTX<(outs RRegf32:$d),
+                     (ins RRegf32:$a),
+                     "sin.approx.f32\t$d, $a",
+                     [(set RRegf32:$d, (fsin RRegf32:$a))]>;
+
+def FSIN64 : InstPTX<(outs RRegf64:$d),
+                     (ins RRegf64:$a),
+                     "sin.approx.f64\t$d, $a",
+                     [(set RRegf64:$d, (fsin RRegf64:$a))]>;
+
+def FCOS32 : InstPTX<(outs RRegf32:$d),
+                     (ins RRegf32:$a),
+                     "cos.approx.f32\t$d, $a",
+                     [(set RRegf32:$d, (fcos RRegf32:$a))]>;
+
+def FCOS64 : InstPTX<(outs RRegf64:$d),
+                     (ins RRegf64:$a),
+                     "cos.approx.f64\t$d, $a",
+                     [(set RRegf64:$d, (fcos RRegf64:$a))]>;
+
+
 ///===- Comparison and Selection Instructions -----------------------------===//
 
 defm SETPEQu32 : PTX_SETP<RRegu32, "u32", i32imm, SETEQ,  "eq">;

Added: llvm/trunk/test/CodeGen/PTX/llvm-intrinsic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PTX/llvm-intrinsic.ll?rev=127578&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/PTX/llvm-intrinsic.ll (added)
+++ llvm/trunk/test/CodeGen/PTX/llvm-intrinsic.ll Mon Mar 14 09:09:33 2011
@@ -0,0 +1,56 @@
+; RUN: llc < %s -march=ptx -mattr=+ptx20,+sm20 | FileCheck %s
+
+define ptx_device float @test_sqrt_f32(float %x) {
+entry:
+; CHECK: sqrt.rn.f32 f0, f1;
+; CHECK-NEXT: ret;
+  %y = call float @llvm.sqrt.f32(float %x)
+  ret float %y
+}
+
+define ptx_device double @test_sqrt_f64(double %x) {
+entry:
+; CHECK: sqrt.rn.f64 fd0, fd1;
+; CHECK-NEXT: ret;
+  %y = call double @llvm.sqrt.f64(double %x)
+  ret double %y
+}
+
+define ptx_device float @test_sin_f32(float %x) {
+entry:
+; CHECK: sin.approx.f32 f0, f1;
+; CHECK-NEXT: ret;
+  %y = call float @llvm.sin.f32(float %x)
+  ret float %y
+}
+
+define ptx_device double @test_sin_f64(double %x) {
+entry:
+; CHECK: sin.approx.f64 fd0, fd1;
+; CHECK-NEXT: ret;
+  %y = call double @llvm.sin.f64(double %x)
+  ret double %y
+}
+
+define ptx_device float @test_cos_f32(float %x) {
+entry:
+; CHECK: cos.approx.f32 f0, f1;
+; CHECK-NEXT: ret;
+  %y = call float @llvm.cos.f32(float %x)
+  ret float %y
+}
+
+define ptx_device double @test_cos_f64(double %x) {
+entry:
+; CHECK: cos.approx.f64 fd0, fd1;
+; CHECK-NEXT: ret;
+  %y = call double @llvm.cos.f64(double %x)
+  ret double %y
+}
+
+declare float  @llvm.sqrt.f32(float)
+declare double @llvm.sqrt.f64(double)
+declare float  @llvm.sin.f32(float)
+declare double @llvm.sin.f64(double)
+declare float  @llvm.cos.f32(float)
+declare double @llvm.cos.f64(double)