[riscv]A little question about get_fpu_register_double

[Jing Peiyang]

I am writing a test about the vfmv_vf instruction as follows:

~~~~~~~~~

diff --git a/test/cctest/test-assembler-riscv64.cc b/test/cctest/test-assembler-riscv64.cc
index b1e018e9a30..29d8d174c80 100644
--- a/test/cctest/test-assembler-riscv64.cc
+++ b/test/cctest/test-assembler-riscv64.cc
@@ -2817,6 +2817,26 @@ UTEST_RVV_VF_VFMERGE_VF_FORM_WITH_RES(float, int32_t, 32,
                                       ((mask >> i) & 0x1) ? rs1_fval : src[i])
 #undef UTEST_RVV_VF_VFMERGE_VF_FORM_WITH_RES
 
+TEST(RISCV_UTEST_vfmv_vf_double_nan) {
+  if (!CpuFeatures::IsSupported(RISCV_SIMD)) return;
+  constexpr uint32_t n = 2;
+  CcTest::InitializeVM();
+
+  // int64_t rs1_fval = 0x7ff7654321fedcba;
+  int64_t rs1_fval = 0x7FF4000000000000;
+  int64_t dst[n] = {0};
+  auto fn = [](MacroAssembler& assm) {
+    __ VU.set(t0, VSew::E64, Vlmul::m1);
+    __ fld(ft0, a0, 0);
+    __ vfmv_vf(v1, ft0);
+    __ vs(v1, a1, 0, VSew::E64);
+  };
+  GenAndRunTest<int64_t, int64_t>((int64_t)&rs1_fval, (int64_t)dst, fn);
+  for (uint32_t i = 0; i < n; i++) {
+    CHECK_EQ(rs1_fval, dst[i]);
+  }
+}
+
 // Tests for vector permutation instructions vector slide instructions
 #define UTEST_RVV_VP_VSLIDE_VI_FORM_WITH_RES(instr_name, type, width, array, \
                                              expect_res)                     \
(END)

~~~~~~~~~

This test has two phenomena in riscv32 and riscv64, riscv32 will normalize 0x7FF4000000000000 to 0x7FFC000000000000, while riscv64 will not change it.

I know that the instruction vfmv_vf should not call the function get_fpu_register_double in the simulator, but should call the function get_fpu_register_Float64.

But I don't quite understand why the function get_fpu_register_double doesn't normalize nan in riscv64?

[Yahan Lu]

Because riscv32 run simulator on x86 but riscv64 run on x64.

X86 normalize 0x7FF4000000000000 to 0x7FFC000000000000 by default.

[Yahan Lu]

https://stackoverflow.com/questions/22816095/signalling-nan-was-corrupted-when-returning-from-x86-function-flds-fstps-of-x87

[Jing Peiyang]

I understand, thank you very much for your answer.