This patch adds a new NVPTX back-end to LLVM which supports code generation for NVIDIA PTX 3.0. This back-end will (eventually) replace the current PTX back-end, while maintaining compatibility with it.

The new target machines are:

nvptx (old ptx32) => 32-bit PTX
nvptx64 (old ptx64) => 64-bit PTX

The sources are based on the internal NVIDIA NVPTX back-end, and
contain more functionality than the current PTX back-end currently
provides.

NV_CONTRIB

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156196 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 72 insertions, 0 deletions

diff --git a/test/CodeGen/NVPTX/arithmetic-fp-sm10.ll b/test/CodeGen/NVPTX/arithmetic-fp-sm10.ll
new file mode 100644
index 0000000000..73c77f56bc
--- /dev/null
+++ b/test/CodeGen/NVPTX/arithmetic-fp-sm10.ll
@@ -0,0 +1,72 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_10 | FileCheck %s
+; RUN: llc < %s -march=nvptx64 -mcpu=sm_10 | FileCheck %s
+
+;; These tests should run for all targets
+
+;;===-- Basic instruction selection tests ---------------------------------===;;
+
+
+;;; f64
+
+define double @fadd_f64(double %a, double %b) {
+; CHECK: add.f64 %fl{{[0-9]+}}, %fl{{[0-9]+}}, %fl{{[0-9]+}}
+; CHECK: ret
+  %ret = fadd double %a, %b
+  ret double %ret
+}
+
+define double @fsub_f64(double %a, double %b) {
+; CHECK: sub.f64 %fl{{[0-9]+}}, %fl{{[0-9]+}}, %fl{{[0-9]+}}
+; CHECK: ret
+  %ret = fsub double %a, %b
+  ret double %ret
+}
+
+define double @fmul_f64(double %a, double %b) {
+; CHECK: mul.f64 %fl{{[0-9]+}}, %fl{{[0-9]+}}, %fl{{[0-9]+}}
+; CHECK: ret
+  %ret = fmul double %a, %b
+  ret double %ret
+}
+
+define double @fdiv_f64(double %a, double %b) {
+; CHECK: div.rn.f64 %fl{{[0-9]+}}, %fl{{[0-9]+}}, %fl{{[0-9]+}}
+; CHECK: ret
+  %ret = fdiv double %a, %b
+  ret double %ret
+}
+
+;; PTX does not have a floating-point rem instruction
+
+
+;;; f32
+
+define float @fadd_f32(float %a, float %b) {
+; CHECK: add.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
+; CHECK: ret
+  %ret = fadd float %a, %b
+  ret float %ret
+}
+
+define float @fsub_f32(float %a, float %b) {
+; CHECK: sub.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
+; CHECK: ret
+  %ret = fsub float %a, %b
+  ret float %ret
+}
+
+define float @fmul_f32(float %a, float %b) {
+; CHECK: mul.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
+; CHECK: ret
+  %ret = fmul float %a, %b
+  ret float %ret
+}
+
+define float @fdiv_f32(float %a, float %b) {
+; CHECK: div.full.f32 %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}
+; CHECK: ret
+  %ret = fdiv float %a, %b
+  ret float %ret
+}
+
+;; PTX does not have a floating-point rem instruction