[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

diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.td b/lib/Target/NVPTX/NVPTXInstrInfo.td
index b23f1e4b458d3a661fec39a8ef211e1fba442c2f..fbcd0e4a358fb9255967393c45557dd51bd8fed6 100644 (file)
--- 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 (file)
index 0000000..7f796e0
--- /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
+}
+