[ARM] Enable DP copy, load and store instructions for FPv4-SP

The FPv4-SP floating-point unit is generally referred to as
single-precision only, but it does have double-precision registers and
load, store and GPR<->DPR move instructions which operate on them.
This patch enables the use of these registers, the main advantage of
which is that we now comply with the AAPCS-VFP calling convention.
This partially reverts r209650, which added some AAPCS-VFP support,
but did not handle return values or alignment of double arguments in
registers.

This patch also adds tests for Thumb2 code generation for
floating-point instructions and intrinsics, which previously only
existed for ARM.

llvm-svn: 216172

1 files changed, 210 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/Thumb2/float-intrinsics-float.ll b/llvm/test/CodeGen/Thumb2/float-intrinsics-float.ll
new file mode 100644
index 00000000000..b29ab35c8f6
--- /dev/null
+++ b/llvm/test/CodeGen/Thumb2/float-intrinsics-float.ll
@@ -0,0 +1,210 @@
+; RUN: llc < %s -mtriple=thumbv7-none-eabi   -mcpu=cortex-m3 | FileCheck %s -check-prefix=CHECK -check-prefix=SOFT -check-prefix=NONE
+; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-m4 | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=SP
+; RUN: llc < %s -mtriple=thumbv7-none-eabihf -mcpu=cortex-a7 | FileCheck %s -check-prefix=CHECK -check-prefix=HARD -check-prefix=DP
+
+declare float     @llvm.sqrt.f32(float %Val)
+define float @sqrt_f(float %a) {
+; CHECK-LABEL: sqrt_f:
+; SOFT: bl sqrtf
+; HARD: vsqrt.f32 s0, s0
+  %1 = call float @llvm.sqrt.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.powi.f32(float %Val, i32 %power)
+define float @powi_f(float %a, i32 %b) {
+; CHECK-LABEL: powi_f:
+; SOFT: bl __powisf2
+; HARD: b __powisf2
+  %1 = call float @llvm.powi.f32(float %a, i32 %b)
+  ret float %1
+}
+
+declare float     @llvm.sin.f32(float %Val)
+define float @sin_f(float %a) {
+; CHECK-LABEL: sin_f:
+; SOFT: bl sinf
+; HARD: b sinf
+  %1 = call float @llvm.sin.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.cos.f32(float %Val)
+define float @cos_f(float %a) {
+; CHECK-LABEL: cos_f:
+; SOFT: bl cosf
+; HARD: b cosf
+  %1 = call float @llvm.cos.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.pow.f32(float %Val, float %power)
+define float @pow_f(float %a, float %b) {
+; CHECK-LABEL: pow_f:
+; SOFT: bl powf
+; HARD: b powf
+  %1 = call float @llvm.pow.f32(float %a, float %b)
+  ret float %1
+}
+
+declare float     @llvm.exp.f32(float %Val)
+define float @exp_f(float %a) {
+; CHECK-LABEL: exp_f:
+; SOFT: bl expf
+; HARD: b expf
+  %1 = call float @llvm.exp.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.exp2.f32(float %Val)
+define float @exp2_f(float %a) {
+; CHECK-LABEL: exp2_f:
+; SOFT: bl exp2f
+; HARD: b exp2f
+  %1 = call float @llvm.exp2.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.log.f32(float %Val)
+define float @log_f(float %a) {
+; CHECK-LABEL: log_f:
+; SOFT: bl logf
+; HARD: b logf
+  %1 = call float @llvm.log.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.log10.f32(float %Val)
+define float @log10_f(float %a) {
+; CHECK-LABEL: log10_f:
+; SOFT: bl log10f
+; HARD: b log10f
+  %1 = call float @llvm.log10.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.log2.f32(float %Val)
+define float @log2_f(float %a) {
+; CHECK-LABEL: log2_f:
+; SOFT: bl log2f
+; HARD: b log2f
+  %1 = call float @llvm.log2.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.fma.f32(float %a, float %b, float %c)
+define float @fma_f(float %a, float %b, float %c) {
+; CHECK-LABEL: fma_f:
+; SOFT: bl fmaf
+; HARD: vfma.f32
+  %1 = call float @llvm.fma.f32(float %a, float %b, float %c)
+  ret float %1
+}
+
+declare float     @llvm.fabs.f32(float %Val)
+define float @abs_f(float %a) {
+; CHECK-LABEL: abs_f:
+; SOFT: bic r0, r0, #-2147483648
+; HARD: vabs.f32
+  %1 = call float @llvm.fabs.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.copysign.f32(float  %Mag, float  %Sgn)
+define float @copysign_f(float %a, float %b) {
+; CHECK-LABEL: copysign_f:
+; NONE: lsrs [[REG:r[0-9]+]], r{{[0-9]+}}, #31
+; NONE: bfi r{{[0-9]+}}, [[REG]], #31, #1
+; SP: lsrs [[REG:r[0-9]+]], r{{[0-9]+}}, #31
+; SP: bfi r{{[0-9]+}}, [[REG]], #31, #1
+; DP: vmov.i32 [[REG:d[0-9]+]], #0x80000000
+; DP: vbsl [[REG]], d
+  %1 = call float @llvm.copysign.f32(float %a, float %b)
+  ret float %1
+}
+
+declare float     @llvm.floor.f32(float %Val)
+define float @floor_f(float %a) {
+; CHECK-LABEL: floor_f:
+; SOFT: bl floorf
+; HARD: b floorf
+  %1 = call float @llvm.floor.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.ceil.f32(float %Val)
+define float @ceil_f(float %a) {
+; CHECK-LABEL: ceil_f:
+; SOFT: bl ceilf
+; HARD: b ceilf
+  %1 = call float @llvm.ceil.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.trunc.f32(float %Val)
+define float @trunc_f(float %a) {
+; CHECK-LABEL: trunc_f:
+; SOFT: bl truncf
+; HARD: b truncf
+  %1 = call float @llvm.trunc.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.rint.f32(float %Val)
+define float @rint_f(float %a) {
+; CHECK-LABEL: rint_f:
+; SOFT: bl rintf
+; HARD: b rintf
+  %1 = call float @llvm.rint.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.nearbyint.f32(float %Val)
+define float @nearbyint_f(float %a) {
+; CHECK-LABEL: nearbyint_f:
+; SOFT: bl nearbyintf
+; HARD: b nearbyintf
+  %1 = call float @llvm.nearbyint.f32(float %a)
+  ret float %1
+}
+
+declare float     @llvm.round.f32(float %Val)
+define float @round_f(float %a) {
+; CHECK-LABEL: round_f:
+; SOFT: bl roundf
+; HARD: b roundf
+  %1 = call float @llvm.round.f32(float %a)
+  ret float %1
+}
+
+; FIXME: why does cortex-m4 use vmla, while cortex-a7 uses vmul+vadd?
+; (these should be equivalent, even the rounding is the same)
+declare float     @llvm.fmuladd.f32(float %a, float %b, float %c)
+define float @fmuladd_f(float %a, float %b, float %c) {
+; CHECK-LABEL: fmuladd_f:
+; SOFT: bl __aeabi_fmul
+; SOFT: bl __aeabi_fadd
+; SP: vmla.f32
+; DP: vmul.f32
+; DP: vadd.f32
+  %1 = call float @llvm.fmuladd.f32(float %a, float %b, float %c)
+  ret float %1
+}
+
+declare i16 @llvm.convert.to.fp16.f32(float %a)
+define i16 @f_to_h(float %a) {
+; CHECK-LABEL: f_to_h:
+; SOFT: bl __gnu_f2h_ieee
+; HARD: vcvtb.f16.f32
+  %1 = call i16 @llvm.convert.to.fp16.f32(float %a)
+  ret i16 %1
+}
+
+declare float @llvm.convert.from.fp16.f32(i16 %a)
+define float @h_to_f(i16 %a) {
+; CHECK-LABEL: h_to_f:
+; SOFT: bl __gnu_h2f_ieee
+; HARD: vcvtb.f32.f16
+  %1 = call float @llvm.convert.from.fp16.f32(i16 %a)
+  ret float %1
+}