1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_20 -verify-machineinstrs | FileCheck %s
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_30 -verify-machineinstrs | FileCheck %s
; RUN: %if ptxas %{ llc < %s -mtriple=nvptx64 -mcpu=sm_20 -verify-machineinstrs | %ptxas-verify %}
; RUN: %if ptxas %{ llc < %s -mtriple=nvptx64 -mcpu=sm_30 -verify-machineinstrs | %ptxas-verify %}
target triple = "nvptx-unknown-cuda"
declare { float, float, float, float } @llvm.nvvm.tex.unified.1d.v4f32.s32(i64, i32)
declare i64 @llvm.nvvm.texsurf.handle.internal.p1(ptr addrspace(1))
define ptx_kernel void @foo(i64 %img, ptr %red, i32 %idx) {
; CHECK-LABEL: foo(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<2>;
; CHECK-NEXT: .reg .f32 %f<5>;
; CHECK-NEXT: .reg .b64 %rd<4>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.u64 %rd1, [foo_param_0];
; CHECK-NEXT: ld.param.u64 %rd2, [foo_param_1];
; CHECK-NEXT: cvta.to.global.u64 %rd3, %rd2;
; CHECK-NEXT: ld.param.u32 %r1, [foo_param_2];
; CHECK-NEXT: tex.1d.v4.f32.s32 {%f1, %f2, %f3, %f4}, [%rd1, {%r1}];
; CHECK-NEXT: st.global.f32 [%rd3], %f1;
; CHECK-NEXT: ret;
%val = tail call { float, float, float, float } @llvm.nvvm.tex.unified.1d.v4f32.s32(i64 %img, i32 %idx)
%ret = extractvalue { float, float, float, float } %val, 0
store float %ret, ptr %red
ret void
}
@tex0 = internal addrspace(1) global i64 0, align 8
define ptx_kernel void @bar(ptr %red, i32 %idx) {
; CHECK-LABEL: bar(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<2>;
; CHECK-NEXT: .reg .f32 %f<5>;
; CHECK-NEXT: .reg .b64 %rd<4>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.u64 %rd1, [bar_param_0];
; CHECK-NEXT: cvta.to.global.u64 %rd2, %rd1;
; CHECK-NEXT: ld.param.u32 %r1, [bar_param_1];
; CHECK-NEXT: tex.1d.v4.f32.s32 {%f1, %f2, %f3, %f4}, [tex0, {%r1}];
; CHECK-NEXT: st.global.f32 [%rd2], %f1;
; CHECK-NEXT: ret;
%texHandle = tail call i64 @llvm.nvvm.texsurf.handle.internal.p1(ptr addrspace(1) @tex0)
%val = tail call { float, float, float, float } @llvm.nvvm.tex.unified.1d.v4f32.s32(i64 %texHandle, i32 %idx)
%ret = extractvalue { float, float, float, float } %val, 0
store float %ret, ptr %red
ret void
}
declare float @texfunc(i64)
define ptx_kernel void @baz(ptr %red, i32 %idx) {
; CHECK-LABEL: baz(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<2>;
; CHECK-NEXT: .reg .f32 %f<8>;
; CHECK-NEXT: .reg .b64 %rd<4>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.u64 %rd1, [baz_param_0];
; CHECK-NEXT: cvta.to.global.u64 %rd2, %rd1;
; CHECK-NEXT: ld.param.u32 %r1, [baz_param_1];
; CHECK-NEXT: mov.u64 %rd3, tex0;
; CHECK-NEXT: tex.1d.v4.f32.s32 {%f1, %f2, %f3, %f4}, [tex0, {%r1}];
; CHECK-NEXT: { // callseq 0, 0
; CHECK-NEXT: .param .b64 param0;
; CHECK-NEXT: st.param.b64 [param0], %rd3;
; CHECK-NEXT: .param .b32 retval0;
; CHECK-NEXT: call.uni (retval0),
; CHECK-NEXT: texfunc,
; CHECK-NEXT: (
; CHECK-NEXT: param0
; CHECK-NEXT: );
; CHECK-NEXT: ld.param.f32 %f5, [retval0];
; CHECK-NEXT: } // callseq 0
; CHECK-NEXT: add.rn.f32 %f7, %f1, %f5;
; CHECK-NEXT: st.global.f32 [%rd2], %f7;
; CHECK-NEXT: ret;
%texHandle = tail call i64 @llvm.nvvm.texsurf.handle.internal.p1(ptr addrspace(1) @tex0)
%val = tail call { float, float, float, float } @llvm.nvvm.tex.unified.1d.v4f32.s32(i64 %texHandle, i32 %idx)
%ret = extractvalue { float, float, float, float } %val, 0
%texcall = tail call float @texfunc(i64 %texHandle)
%ret2 = fadd float %ret, %texcall
store float %ret2, ptr %red
ret void
}
!nvvm.annotations = !{!1}
!1 = !{ptr addrspace(1) @tex0, !"texture", i32 1}
|
