summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJustin Holewinski <jholewinski@nvidia.com>2014-01-21 14:40:05 +0000
committerJustin Holewinski <jholewinski@nvidia.com>2014-01-21 14:40:05 +0000
commitc93ed1707a00ee7d0d26d6a716d566f9ec19a8a2 (patch)
tree0da3a81f6ad1c6e724b25079cb47bd2a86aece3c
parentdc5734786aa9260d5995a098d38aa4deb13e83c0 (diff)
downloadllvm-c93ed1707a00ee7d0d26d6a716d566f9ec19a8a2.tar.gz
llvm-c93ed1707a00ee7d0d26d6a716d566f9ec19a8a2.tar.bz2
llvm-c93ed1707a00ee7d0d26d6a716d566f9ec19a8a2.tar.xz
[NVPTX] Add missing patterns for div.approx with immediate denominator
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199746 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Target/NVPTX/NVPTXInstrInfo.td12
-rw-r--r--test/CodeGen/NVPTX/div-ri.ll8
2 files changed, 20 insertions, 0 deletions
diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.td b/lib/Target/NVPTX/NVPTXInstrInfo.td
index b23f1e4b45..fbcd0e4a35 100644
--- a/lib/Target/NVPTX/NVPTXInstrInfo.td
+++ b/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -689,12 +689,24 @@ def FDIV32approxrr_ftz : NVPTXInst<(outs Float32Regs:$dst),
[(set Float32Regs:$dst,
(fdiv Float32Regs:$a, Float32Regs:$b))]>,
Requires<[do_DIVF32_APPROX, doF32FTZ]>;
+def FDIV32approxri_ftz : NVPTXInst<(outs Float32Regs:$dst),
+ (ins Float32Regs:$a, f32imm:$b),
+ "div.approx.ftz.f32 \t$dst, $a, $b;",
+ [(set Float32Regs:$dst,
+ (fdiv Float32Regs:$a, fpimm:$b))]>,
+ Requires<[do_DIVF32_APPROX, doF32FTZ]>;
def FDIV32approxrr : NVPTXInst<(outs Float32Regs:$dst),
(ins Float32Regs:$a, Float32Regs:$b),
"div.approx.f32 \t$dst, $a, $b;",
[(set Float32Regs:$dst,
(fdiv Float32Regs:$a, Float32Regs:$b))]>,
Requires<[do_DIVF32_APPROX]>;
+def FDIV32approxri : NVPTXInst<(outs Float32Regs:$dst),
+ (ins Float32Regs:$a, f32imm:$b),
+ "div.approx.f32 \t$dst, $a, $b;",
+ [(set Float32Regs:$dst,
+ (fdiv Float32Regs:$a, fpimm:$b))]>,
+ Requires<[do_DIVF32_APPROX]>;
//
// F32 Semi-accurate reciprocal
//
diff --git a/test/CodeGen/NVPTX/div-ri.ll b/test/CodeGen/NVPTX/div-ri.ll
new file mode 100644
index 0000000000..7f796e0239
--- /dev/null
+++ b/test/CodeGen/NVPTX/div-ri.ll
@@ -0,0 +1,8 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 -nvptx-prec-divf32=0 | FileCheck %s
+
+define float @foo(float %a) {
+; CHECK: div.approx.f32
+ %div = fdiv float %a, 13.0
+ ret float %div
+}
+