2015-02-11 22:26:46 +08:00
|
|
|
; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
|
2014-12-22 00:48:42 +08:00
|
|
|
; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
|
2014-07-20 14:11:02 +08:00
|
|
|
|
2014-12-22 00:48:42 +08:00
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_0:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT:buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
|
|
|
|
; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W
|
|
|
|
; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1
|
|
|
|
define void @sext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, 0
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_0:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
|
|
|
|
; EG: SETNE_INT * [[CMP:T[0-9]+]].[[CMPCHAN:[XYZW]]], KC0[2].Z, KC0[2].W
|
|
|
|
; EG: AND_INT T{{[0-9]+.[XYZW]}}, PS, 1
|
|
|
|
define void @sext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, 0
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; This really folds away to false
|
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: v_cmp_eq_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, -1, vcc
|
2015-03-24 02:45:30 +08:00
|
|
|
; GCN-NEXT: v_cmp_eq_i32_e32 vcc, 1, [[TMP]]{{$}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, 1,
|
|
|
|
; GCN-NEXT: buffer_store_byte [[TMP]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @sext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, 1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; This really folds away to true
|
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, -1, vcc
|
2015-03-24 02:45:30 +08:00
|
|
|
; GCN-NEXT: v_cmp_ne_i32_e32 vcc, 1, [[TMP]]{{$}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[TMP:v[0-9]+]], 0, 1,
|
|
|
|
; GCN-NEXT: buffer_store_byte [[TMP]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @sext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
2014-07-20 14:11:02 +08:00
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
2014-12-22 00:48:42 +08:00
|
|
|
%icmp1 = icmp ne i32 %ext, 1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2016-01-20 08:13:22 +08:00
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_eq_neg1:
|
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_eq_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
|
|
|
define void @sext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, -1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_neg1:
|
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_eq_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
|
|
|
define void @sext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, -1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2014-12-22 00:48:42 +08:00
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_0:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_eq_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, 0
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_0:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_ne_0(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
2014-07-20 14:11:02 +08:00
|
|
|
%icmp1 = icmp ne i32 %ext, 0
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
2014-12-22 00:48:42 +08:00
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_eq_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_eq_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, 1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NOT: v_cmp
|
|
|
|
; GCN: v_cmp_eq_i32_e32 vcc,
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_ne_1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, 1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2016-01-20 08:13:22 +08:00
|
|
|
; Reduces to false:
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_neg1:
|
|
|
|
; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 0{{$}}
|
|
|
|
; GCN: buffer_store_byte [[TMP]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
|
|
|
define void @zext_bool_icmp_eq_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp eq i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, -1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Reduces to true:
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_neg1:
|
|
|
|
; GCN: v_mov_b32_e32 [[TMP:v[0-9]+]], 1{{$}}
|
|
|
|
; GCN: buffer_store_byte [[TMP]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
|
|
|
define void @zext_bool_icmp_ne_neg1(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, -1
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
2014-12-22 00:48:42 +08:00
|
|
|
; FUNC-LABEL: {{^}}sext_bool_icmp_ne_k:
|
|
|
|
; SI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
|
|
|
|
; SI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xc
|
2015-02-11 22:26:46 +08:00
|
|
|
; VI-DAG: s_load_dword [[A:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
|
|
|
|
; VI-DAG: s_load_dword [[B:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x30
|
|
|
|
; GCN: v_mov_b32_e32 [[VB:v[0-9]+]], [[B]]
|
2015-03-24 02:45:30 +08:00
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc, 2, [[VB]]{{$}}
|
|
|
|
; GCN: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: buffer_store_byte
|
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @sext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = sext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, 2
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}cmp_zext_k_i8max:
|
2015-11-07 05:58:37 +08:00
|
|
|
; SI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0xb
|
|
|
|
; VI: s_load_dword [[VALUE:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
|
|
|
|
; GCN: s_movk_i32 [[K255:s[0-9]+]], 0xff
|
|
|
|
; GCN: s_and_b32 [[B:s[0-9]+]], [[VALUE]], [[K255]]
|
|
|
|
; GCN: v_mov_b32_e32 [[VK255:v[0-9]+]], [[K255]]
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc, [[B]], [[VK255]]
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
AMDGPU/SI: Better handle s_wait insertion
We can wait on either VM, EXP or LGKM.
The waits are independent.
Without this patch, a wait inserted because of one of them
would also wait for all the previous others.
This patch makes s_wait only wait for the ones we need for the next
instruction.
Here's an example of subtle perf reduction this patch solves:
This is without the patch:
buffer_load_format_xyzw v[8:11], v0, s[44:47], 0 idxen
buffer_load_format_xyzw v[12:15], v0, s[48:51], 0 idxen
s_load_dwordx4 s[44:47], s[8:9], 0xc
s_waitcnt lgkmcnt(0)
buffer_load_format_xyzw v[16:19], v0, s[52:55], 0 idxen
s_load_dwordx4 s[48:51], s[8:9], 0x10
s_waitcnt vmcnt(1)
buffer_load_format_xyzw v[20:23], v0, s[44:47], 0 idxen
The s_waitcnt vmcnt(1) is useless.
The reason it is added is because the last
buffer_load_format_xyzw needs s[44:47], which was issued
by the first s_load_dwordx4. It waits for all VM
before that call to have finished.
Internally after every instruction, 3 counters (for VM, EXP and LGTM)
are updated after every instruction. For example buffer_load_format_xyzw
will
increase the VM counter, and s_load_dwordx4 the LGKM one.
Without the patch, for every defined register,
the current 3 counters are stored, and are used to know
how long to wait when an instruction needs the register.
Because of that, the s[44:47] counter includes that to use the register
you need to wait for the previous buffer_load_format_xyzw.
Instead this patch stores only the counters that matter for the
register,
and puts zero for the other ones, since we don't need any wait for them.
Patch by: Axel Davy
Differential Revision: http://reviews.llvm.org/D11883
llvm-svn: 245755
2015-08-22 06:47:27 +08:00
|
|
|
; GCN: buffer_store_byte [[RESULT]]
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @cmp_zext_k_i8max(i1 addrspace(1)* %out, i8 %b) nounwind {
|
|
|
|
%b.ext = zext i8 %b to i32
|
|
|
|
%icmp0 = icmp ne i32 %b.ext, 255
|
|
|
|
store i1 %icmp0, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}cmp_sext_k_neg1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: buffer_load_sbyte [[B:v[0-9]+]]
|
2015-03-24 02:45:30 +08:00
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc, -1, [[B]]{{$}}
|
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @cmp_sext_k_neg1(i1 addrspace(1)* %out, i8 addrspace(1)* %b.ptr) nounwind {
|
2015-02-28 05:17:42 +08:00
|
|
|
%b = load i8, i8 addrspace(1)* %b.ptr
|
2014-12-22 00:48:42 +08:00
|
|
|
%b.ext = sext i8 %b to i32
|
|
|
|
%icmp0 = icmp ne i32 %b.ext, -1
|
|
|
|
store i1 %icmp0, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_sext_arg:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_load_dword [[B:s[0-9]+]]
|
2015-03-24 02:45:30 +08:00
|
|
|
; GCN: v_cmp_ne_i32_e64 [[CMP:s\[[0-9]+:[0-9]+\]]], -1, [[B]]
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, [[CMP]]
|
|
|
|
; GCN-NEXT: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @cmp_sext_k_neg1_i8_sext_arg(i1 addrspace(1)* %out, i8 signext %b) nounwind {
|
|
|
|
%b.ext = sext i8 %b to i32
|
|
|
|
%icmp0 = icmp ne i32 %b.ext, -1
|
|
|
|
store i1 %icmp0, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FIXME: This ends up doing a buffer_load_ubyte, and and compare to
|
|
|
|
; 255. Seems to be because of ordering problems when not allowing load widths to be reduced.
|
|
|
|
; Should do a buffer_load_sbyte and compare with -1
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}cmp_sext_k_neg1_i8_arg:
|
2015-11-07 05:58:37 +08:00
|
|
|
; SI: s_load_dword [[VAL:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
|
|
|
|
; VI: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, 0x2c
|
|
|
|
; GCN: s_movk_i32 [[K:s[0-9]+]], 0xff
|
|
|
|
; GCN: s_and_b32 [[B:s[0-9]+]], [[VAL]], [[K]]
|
|
|
|
; GCN: v_mov_b32_e32 [[VK:v[0-9]+]], [[K]]
|
|
|
|
; GCN: v_cmp_ne_i32_e32 vcc, [[B]], [[VK]]{{$}}
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN-NEXT: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], 0, 1, vcc
|
AMDGPU/SI: Better handle s_wait insertion
We can wait on either VM, EXP or LGKM.
The waits are independent.
Without this patch, a wait inserted because of one of them
would also wait for all the previous others.
This patch makes s_wait only wait for the ones we need for the next
instruction.
Here's an example of subtle perf reduction this patch solves:
This is without the patch:
buffer_load_format_xyzw v[8:11], v0, s[44:47], 0 idxen
buffer_load_format_xyzw v[12:15], v0, s[48:51], 0 idxen
s_load_dwordx4 s[44:47], s[8:9], 0xc
s_waitcnt lgkmcnt(0)
buffer_load_format_xyzw v[16:19], v0, s[52:55], 0 idxen
s_load_dwordx4 s[48:51], s[8:9], 0x10
s_waitcnt vmcnt(1)
buffer_load_format_xyzw v[20:23], v0, s[44:47], 0 idxen
The s_waitcnt vmcnt(1) is useless.
The reason it is added is because the last
buffer_load_format_xyzw needs s[44:47], which was issued
by the first s_load_dwordx4. It waits for all VM
before that call to have finished.
Internally after every instruction, 3 counters (for VM, EXP and LGTM)
are updated after every instruction. For example buffer_load_format_xyzw
will
increase the VM counter, and s_load_dwordx4 the LGKM one.
Without the patch, for every defined register,
the current 3 counters are stored, and are used to know
how long to wait when an instruction needs the register.
Because of that, the s[44:47] counter includes that to use the register
you need to wait for the previous buffer_load_format_xyzw.
Instead this patch stores only the counters that matter for the
register,
and puts zero for the other ones, since we don't need any wait for them.
Patch by: Axel Davy
Differential Revision: http://reviews.llvm.org/D11883
llvm-svn: 245755
2015-08-22 06:47:27 +08:00
|
|
|
; GCN: buffer_store_byte [[RESULT]]
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @cmp_sext_k_neg1_i8_arg(i1 addrspace(1)* %out, i8 %b) nounwind {
|
|
|
|
%b.ext = sext i8 %b to i32
|
|
|
|
%icmp0 = icmp ne i32 %b.ext, -1
|
|
|
|
store i1 %icmp0, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}cmp_zext_k_neg1:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
|
|
|
|
; GCN: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @cmp_zext_k_neg1(i1 addrspace(1)* %out, i8 %b) nounwind {
|
|
|
|
%b.ext = zext i8 %b to i32
|
|
|
|
%icmp0 = icmp ne i32 %b.ext, -1
|
|
|
|
store i1 %icmp0, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_ne_k:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 1{{$}}
|
|
|
|
; GCN: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_ne_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp ne i32 %ext, 2
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; FUNC-LABEL: {{^}}zext_bool_icmp_eq_k:
|
2015-02-11 22:26:46 +08:00
|
|
|
; GCN: v_mov_b32_e32 [[RESULT:v[0-9]+]], 0{{$}}
|
|
|
|
; GCN: buffer_store_byte [[RESULT]]
|
|
|
|
; GCN-NEXT: s_endpgm
|
2014-12-22 00:48:42 +08:00
|
|
|
define void @zext_bool_icmp_eq_k(i1 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
|
|
|
|
%icmp0 = icmp ne i32 %a, %b
|
|
|
|
%ext = zext i1 %icmp0 to i32
|
|
|
|
%icmp1 = icmp eq i32 %ext, 2
|
|
|
|
store i1 %icmp1, i1 addrspace(1)* %out
|
|
|
|
ret void
|
|
|
|
}
|