[PATCH 1/1] R600: Fix FROUND

[Tom Stellard]

On Thu, Sep 04, 2014 at 04:52:47PM -0400, Jan Vesely wrote:
> round halfway cases away from zero
> 

LGTM.

> Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
> ---
> 
> Fixes 'round' piglit
> 
>  lib/Target/R600/EvergreenInstructions.td |  2 +-
>  lib/Target/R600/R600Instructions.td      |  9 ++++++---
>  test/CodeGen/R600/llvm.round.ll          | 14 ++++++++------
>  3 files changed, 15 insertions(+), 10 deletions(-)
> 
> diff --git a/lib/Target/R600/EvergreenInstructions.td b/lib/Target/R600/EvergreenInstructions.td
> index a83567a..ea3f550 100644
> --- a/lib/Target/R600/EvergreenInstructions.td
> +++ b/lib/Target/R600/EvergreenInstructions.td
> @@ -537,7 +537,7 @@ def : Pat<(fp_to_uint f32:$src0), (FLT_TO_UINT_eg (TRUNC $src0))>;
>  // SHA-256 Patterns
>  def : SHA256MaPattern <BFI_INT_eg, XOR_INT>;
>  
> -def : FROUNDPat <CNDGE_eg>;
> +def : FROUNDPat <CNDGE_eg, CNDGT_eg>;
>  
>  def EG_ExportSwz : ExportSwzInst {
>    let Word1{19-16} = 0; // BURST_COUNT
> diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
> index 671f923..a883c83 100644
> --- a/lib/Target/R600/R600Instructions.td
> +++ b/lib/Target/R600/R600Instructions.td
> @@ -1141,9 +1141,12 @@ class TGSI_LIT_Z_Common <InstR600 mul_lit, InstR600 log_clamped, InstR600 exp_ie
>  >;
>  
>  // FROUND pattern
> -class FROUNDPat<Instruction CNDGE> : Pat <
> +class FROUNDPat<Instruction CNDGE, Instruction CNDGT> : Pat <
>    (AMDGPUround f32:$x),
> -  (CNDGE (ADD (FNEG_R600 (f32 HALF)), (FRACT $x)), (CEIL $x), (FLOOR $x))
> +  (CNDGE $x,
> +  (CNDGE (ADD (FNEG_R600 (f32 HALF)), (FRACT $x)), (CEIL $x), (FLOOR $x)),
> +  (CNDGT (ADD (FNEG_R600 (f32 HALF)), (FRACT $x)), (CEIL $x), (FLOOR $x))
> +  )
>  >;
>  
>  
> @@ -1190,7 +1193,7 @@ let Predicates = [isR600] in {
>    def : Pat<(fsqrt f32:$src), (MUL $src, (RECIPSQRT_CLAMPED_r600 $src))>;
>    defm : RsqPat<RECIPSQRT_IEEE_r600, f32>;
>  
> -  def : FROUNDPat <CNDGE_r600>;
> +  def : FROUNDPat <CNDGE_r600, CNDGT_r600>;
>  
>    def R600_ExportSwz : ExportSwzInst {
>      let Word1{20-17} = 0; // BURST_COUNT
> diff --git a/test/CodeGen/R600/llvm.round.ll b/test/CodeGen/R600/llvm.round.ll
> index e06d45d..5700427 100644
> --- a/test/CodeGen/R600/llvm.round.ll
> +++ b/test/CodeGen/R600/llvm.round.ll
> @@ -1,11 +1,13 @@
> -; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600 --check-prefix=FUNC
> +; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s --check-prefix=R600 --check-prefix=FUNC
>  
>  ; FUNC-LABEL: @f32
> -; R600: FRACT
> -; R600-DAG: ADD
> -; R600-DAG: CEIL
> -; R600-DAG: FLOOR
> -; R600: CNDGE
> +; R600: FRACT {{.*}}, [[ARG:KC[0-9]\[[0-9]+\]\.[XYZW]]]
> +; R600-DAG: ADD  {{.*}}, -0.5
> +; R600-DAG: CEIL {{.*}} [[ARG]]
> +; R600-DAG: FLOOR {{.*}} [[ARG]]
> +; R600-DAG: CNDGE
> +; R600-DAG: CNDGT
> +; R600: CNDGE {{[^,]+}}, [[ARG]]
>  define void @f32(float addrspace(1)* %out, float %in) {
>  entry:
>    %0 = call float @llvm.round.f32(float %in)
> -- 
> 1.9.3
>