Skip to content

Commit

Permalink
R600/SI: mark most intrinsics as readnone v2
Browse files Browse the repository at this point in the history
They read from constant register space anyway.

v2: fix lit tests

Signed-off-by: Christian König <[email protected]>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178020 91177308-0d34-0410-b5e6-96231b3b80d8
  • Loading branch information
ChristianKoenigAMD committed Mar 26, 2013
1 parent 03cd75e commit f623008
Show file tree
Hide file tree
Showing 3 changed files with 89 additions and 54 deletions.
10 changes: 5 additions & 5 deletions lib/Target/R600/SIIntrinsics.td
Original file line number Diff line number Diff line change
Expand Up @@ -16,19 +16,19 @@ let TargetPrefix = "SI", isTarget = 1 in {

def int_SI_packf16 : Intrinsic <[llvm_i32_ty], [llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
def int_SI_export : Intrinsic <[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_float_ty, llvm_float_ty, llvm_float_ty, llvm_float_ty], []>;
def int_SI_load_const : Intrinsic <[llvm_float_ty], [llvm_v16i8_ty, llvm_i32_ty], [IntrReadMem]>;
def int_SI_vs_load_input : Intrinsic <[llvm_v4f32_ty], [llvm_v16i8_ty, llvm_i16_ty, llvm_i32_ty], [IntrReadMem]> ;
def int_SI_load_const : Intrinsic <[llvm_float_ty], [llvm_v16i8_ty, llvm_i32_ty], [IntrNoMem]>;
def int_SI_vs_load_input : Intrinsic <[llvm_v4f32_ty], [llvm_v16i8_ty, llvm_i16_ty, llvm_i32_ty], [IntrNoMem]> ;

class Sample : Intrinsic <[llvm_v4f32_ty], [llvm_i32_ty, llvm_anyvector_ty, llvm_v32i8_ty, llvm_v16i8_ty, llvm_i32_ty], [IntrReadMem]>;
class Sample : Intrinsic <[llvm_v4f32_ty], [llvm_i32_ty, llvm_anyvector_ty, llvm_v32i8_ty, llvm_v16i8_ty, llvm_i32_ty], [IntrNoMem]>;

def int_SI_sample : Sample;
def int_SI_sampleb : Sample;
def int_SI_samplel : Sample;

/* Interpolation Intrinsics */

def int_SI_fs_constant : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrReadMem]>;
def int_SI_fs_interp : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_v2i32_ty], [IntrReadMem]>;
def int_SI_fs_constant : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
def int_SI_fs_interp : Intrinsic <[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_v2i32_ty], [IntrNoMem]>;

/* Control flow Intrinsics */

Expand Down
2 changes: 1 addition & 1 deletion test/CodeGen/R600/llvm.SI.fs.interp.constant.ll
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ main_body:

declare void @llvm.AMDGPU.shader.type(i32)

declare float @llvm.SI.fs.constant(i32, i32, i32) readonly
declare float @llvm.SI.fs.constant(i32, i32, i32) readnone

declare i32 @llvm.SI.packf16(float, float) readnone

Expand Down
131 changes: 83 additions & 48 deletions test/CodeGen/R600/llvm.SI.sample.ll
Original file line number Diff line number Diff line change
@@ -1,71 +1,106 @@
;RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s

;CHECK: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE_C
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK-NEXT: S_WAITCNT 1904
;CHECK-NEXT: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE_C
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE
;CHECK: IMAGE_SAMPLE

define void @test() {
%res1 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
define void @test(i32 %a1, i32 %a2, i32 %a3, i32 %a4) {
%v1 = insertelement <4 x i32> undef, i32 %a1, i32 0
%v2 = insertelement <4 x i32> undef, i32 %a1, i32 1
%v3 = insertelement <4 x i32> undef, i32 %a1, i32 2
%v4 = insertelement <4 x i32> undef, i32 %a1, i32 3
%v5 = insertelement <4 x i32> undef, i32 %a2, i32 0
%v6 = insertelement <4 x i32> undef, i32 %a2, i32 1
%v7 = insertelement <4 x i32> undef, i32 %a2, i32 2
%v8 = insertelement <4 x i32> undef, i32 %a2, i32 3
%v9 = insertelement <4 x i32> undef, i32 %a3, i32 0
%v10 = insertelement <4 x i32> undef, i32 %a3, i32 1
%v11 = insertelement <4 x i32> undef, i32 %a3, i32 2
%v12 = insertelement <4 x i32> undef, i32 %a3, i32 3
%v13 = insertelement <4 x i32> undef, i32 %a4, i32 0
%v14 = insertelement <4 x i32> undef, i32 %a4, i32 1
%v15 = insertelement <4 x i32> undef, i32 %a4, i32 2
%v16 = insertelement <4 x i32> undef, i32 %a4, i32 3
%res1 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v1,
<8 x i32> undef, <4 x i32> undef, i32 1)
%res2 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res2 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v2,
<8 x i32> undef, <4 x i32> undef, i32 2)
%res3 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res3 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v3,
<8 x i32> undef, <4 x i32> undef, i32 3)
%res4 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res4 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v4,
<8 x i32> undef, <4 x i32> undef, i32 4)
%res5 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res5 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v5,
<8 x i32> undef, <4 x i32> undef, i32 5)
%res6 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res6 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v6,
<8 x i32> undef, <4 x i32> undef, i32 6)
%res7 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res7 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v7,
<8 x i32> undef, <4 x i32> undef, i32 7)
%res8 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res8 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v8,
<8 x i32> undef, <4 x i32> undef, i32 8)
%res9 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res9 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v9,
<8 x i32> undef, <4 x i32> undef, i32 9)
%res10 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res10 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v10,
<8 x i32> undef, <4 x i32> undef, i32 10)
%res11 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res11 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v11,
<8 x i32> undef, <4 x i32> undef, i32 11)
%res12 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res12 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v12,
<8 x i32> undef, <4 x i32> undef, i32 12)
%res13 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res13 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v13,
<8 x i32> undef, <4 x i32> undef, i32 13)
%res14 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res14 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v14,
<8 x i32> undef, <4 x i32> undef, i32 14)
%res15 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res15 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v15,
<8 x i32> undef, <4 x i32> undef, i32 15)
%res16 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> undef,
%res16 = call <4 x float> @llvm.SI.sample.(i32 15, <4 x i32> %v16,
<8 x i32> undef, <4 x i32> undef, i32 16)
%e1 = extractelement <4 x float> %res1, i32 0
%e2 = extractelement <4 x float> %res2, i32 0
%e3 = extractelement <4 x float> %res3, i32 0
%e4 = extractelement <4 x float> %res4, i32 0
%e5 = extractelement <4 x float> %res5, i32 0
%e6 = extractelement <4 x float> %res6, i32 0
%e7 = extractelement <4 x float> %res7, i32 0
%e8 = extractelement <4 x float> %res8, i32 0
%e9 = extractelement <4 x float> %res9, i32 0
%e10 = extractelement <4 x float> %res10, i32 0
%e11 = extractelement <4 x float> %res11, i32 0
%e12 = extractelement <4 x float> %res12, i32 0
%e13 = extractelement <4 x float> %res13, i32 0
%e14 = extractelement <4 x float> %res14, i32 0
%e15 = extractelement <4 x float> %res15, i32 0
%e16 = extractelement <4 x float> %res16, i32 0
%s1 = fadd float %e1, %e2
%s2 = fadd float %s1, %e3
%s3 = fadd float %s2, %e4
%s4 = fadd float %s3, %e5
%s5 = fadd float %s4, %e6
%s6 = fadd float %s5, %e7
%s7 = fadd float %s6, %e8
%s8 = fadd float %s7, %e9
%s9 = fadd float %s8, %e10
%s10 = fadd float %s9, %e11
%s11 = fadd float %s10, %e12
%s12 = fadd float %s11, %e13
%s13 = fadd float %s12, %e14
%s14 = fadd float %s13, %e15
%s15 = fadd float %s14, %e16
call void @llvm.SI.export(i32 15, i32 0, i32 1, i32 12, i32 0, float %s15, float %s15, float %s15, float %s15)
ret void
}

declare <4 x float> @llvm.SI.sample.(i32, <4 x i32>, <8 x i32>, <4 x i32>, i32)
declare <4 x float> @llvm.SI.sample.(i32, <4 x i32>, <8 x i32>, <4 x i32>, i32) readnone

declare void @llvm.SI.export(i32, i32, i32, i32, i32, float, float, float, float)

0 comments on commit f623008

Please sign in to comment.