summaryrefslogtreecommitdiffstats
path: root/test/CodeGen/R600/load.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/R600/load.ll')
-rw-r--r--test/CodeGen/R600/load.ll597
1 files changed, 299 insertions, 298 deletions
diff --git a/test/CodeGen/R600/load.ll b/test/CodeGen/R600/load.ll
index 62d3063..b71b7cb 100644
--- a/test/CodeGen/R600/load.ll
+++ b/test/CodeGen/R600/load.ll
@@ -1,6 +1,7 @@
-; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=R600-CHECK --check-prefix=FUNC %s
-; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck --check-prefix=R600-CHECK --check-prefix=FUNC %s
-; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=R600 --check-prefix=FUNC %s
+; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck --check-prefix=R600 --check-prefix=FUNC %s
+; RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
+; RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
;===------------------------------------------------------------------------===;
; GLOBAL ADDRESS SPACE
@@ -8,9 +9,9 @@
; Load an i8 value from the global address space.
; FUNC-LABEL: {{^}}load_i8:
-; R600-CHECK: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
+; R600: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ubyte v{{[0-9]+}},
+; SI: buffer_load_ubyte v{{[0-9]+}},
define void @load_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) {
%1 = load i8 addrspace(1)* %in
%2 = zext i8 %1 to i32
@@ -19,12 +20,12 @@ define void @load_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) {
}
; FUNC-LABEL: {{^}}load_i8_sext:
-; R600-CHECK: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]]
-; R600-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
-; R600-CHECK: 24
-; R600-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
-; R600-CHECK: 24
-; SI-CHECK: buffer_load_sbyte
+; R600: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]]
+; R600: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
+; R600: 24
+; R600: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
+; R600: 24
+; SI: buffer_load_sbyte
define void @load_i8_sext(i32 addrspace(1)* %out, i8 addrspace(1)* %in) {
entry:
%0 = load i8 addrspace(1)* %in
@@ -34,10 +35,10 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i8:
-; R600-CHECK: VTX_READ_8
-; R600-CHECK: VTX_READ_8
-; SI-CHECK: buffer_load_ubyte
-; SI-CHECK: buffer_load_ubyte
+; R600: VTX_READ_8
+; R600: VTX_READ_8
+; SI: buffer_load_ubyte
+; SI: buffer_load_ubyte
define void @load_v2i8(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) {
entry:
%0 = load <2 x i8> addrspace(1)* %in
@@ -47,18 +48,18 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i8_sext:
-; R600-CHECK-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
-; R600-CHECK-DAG: 24
-; SI-CHECK: buffer_load_sbyte
-; SI-CHECK: buffer_load_sbyte
+; R600-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
+; R600-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
+; R600-DAG: 24
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
+; R600-DAG: 24
+; SI: buffer_load_sbyte
+; SI: buffer_load_sbyte
define void @load_v2i8_sext(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(1)* %in) {
entry:
%0 = load <2 x i8> addrspace(1)* %in
@@ -68,14 +69,14 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i8:
-; R600-CHECK: VTX_READ_8
-; R600-CHECK: VTX_READ_8
-; R600-CHECK: VTX_READ_8
-; R600-CHECK: VTX_READ_8
-; SI-CHECK: buffer_load_ubyte
-; SI-CHECK: buffer_load_ubyte
-; SI-CHECK: buffer_load_ubyte
-; SI-CHECK: buffer_load_ubyte
+; R600: VTX_READ_8
+; R600: VTX_READ_8
+; R600: VTX_READ_8
+; R600: VTX_READ_8
+; SI: buffer_load_ubyte
+; SI: buffer_load_ubyte
+; SI: buffer_load_ubyte
+; SI: buffer_load_ubyte
define void @load_v4i8(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) {
entry:
%0 = load <4 x i8> addrspace(1)* %in
@@ -85,30 +86,30 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i8_sext:
-; R600-CHECK-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: VTX_READ_8 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]]
-; R600-CHECK-DAG: VTX_READ_8 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]]
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Z_CHAN:[XYZW]]], [[DST_Z]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Z_CHAN]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_W_CHAN:[XYZW]]], [[DST_W]]
-; R600-CHECK-DAG: 24
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_W_CHAN]]
-; R600-CHECK-DAG: 24
-; SI-CHECK: buffer_load_sbyte
-; SI-CHECK: buffer_load_sbyte
-; SI-CHECK: buffer_load_sbyte
-; SI-CHECK: buffer_load_sbyte
+; R600-DAG: VTX_READ_8 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
+; R600-DAG: VTX_READ_8 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
+; R600-DAG: VTX_READ_8 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]]
+; R600-DAG: VTX_READ_8 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]]
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
+; R600-DAG: 24
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
+; R600-DAG: 24
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Z_CHAN:[XYZW]]], [[DST_Z]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Z_CHAN]]
+; R600-DAG: 24
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_W_CHAN:[XYZW]]], [[DST_W]]
+; R600-DAG: 24
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_W_CHAN]]
+; R600-DAG: 24
+; SI: buffer_load_sbyte
+; SI: buffer_load_sbyte
+; SI: buffer_load_sbyte
+; SI: buffer_load_sbyte
define void @load_v4i8_sext(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(1)* %in) {
entry:
%0 = load <4 x i8> addrspace(1)* %in
@@ -119,8 +120,8 @@ entry:
; Load an i16 value from the global address space.
; FUNC-LABEL: {{^}}load_i16:
-; R600-CHECK: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ushort
+; R600: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
+; SI: buffer_load_ushort
define void @load_i16(i32 addrspace(1)* %out, i16 addrspace(1)* %in) {
entry:
%0 = load i16 addrspace(1)* %in
@@ -130,12 +131,12 @@ entry:
}
; FUNC-LABEL: {{^}}load_i16_sext:
-; R600-CHECK: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]]
-; R600-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
-; R600-CHECK: 16
-; R600-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
-; R600-CHECK: 16
-; SI-CHECK: buffer_load_sshort
+; R600: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]]
+; R600: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
+; R600: 16
+; R600: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
+; R600: 16
+; SI: buffer_load_sshort
define void @load_i16_sext(i32 addrspace(1)* %out, i16 addrspace(1)* %in) {
entry:
%0 = load i16 addrspace(1)* %in
@@ -145,10 +146,10 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i16:
-; R600-CHECK: VTX_READ_16
-; R600-CHECK: VTX_READ_16
-; SI-CHECK: buffer_load_ushort
-; SI-CHECK: buffer_load_ushort
+; R600: VTX_READ_16
+; R600: VTX_READ_16
+; SI: buffer_load_ushort
+; SI: buffer_load_ushort
define void @load_v2i16(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) {
entry:
%0 = load <2 x i16> addrspace(1)* %in
@@ -158,18 +159,18 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i16_sext:
-; R600-CHECK-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
-; R600-CHECK-DAG: 16
-; SI-CHECK: buffer_load_sshort
-; SI-CHECK: buffer_load_sshort
+; R600-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
+; R600-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
+; R600-DAG: 16
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
+; R600-DAG: 16
+; SI: buffer_load_sshort
+; SI: buffer_load_sshort
define void @load_v2i16_sext(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) {
entry:
%0 = load <2 x i16> addrspace(1)* %in
@@ -179,14 +180,14 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i16:
-; R600-CHECK: VTX_READ_16
-; R600-CHECK: VTX_READ_16
-; R600-CHECK: VTX_READ_16
-; R600-CHECK: VTX_READ_16
-; SI-CHECK: buffer_load_ushort
-; SI-CHECK: buffer_load_ushort
-; SI-CHECK: buffer_load_ushort
-; SI-CHECK: buffer_load_ushort
+; R600: VTX_READ_16
+; R600: VTX_READ_16
+; R600: VTX_READ_16
+; R600: VTX_READ_16
+; SI: buffer_load_ushort
+; SI: buffer_load_ushort
+; SI: buffer_load_ushort
+; SI: buffer_load_ushort
define void @load_v4i16(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) {
entry:
%0 = load <4 x i16> addrspace(1)* %in
@@ -196,30 +197,30 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i16_sext:
-; R600-CHECK-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: VTX_READ_16 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]]
-; R600-CHECK-DAG: VTX_READ_16 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]]
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Z_CHAN:[XYZW]]], [[DST_Z]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Z_CHAN]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_W_CHAN:[XYZW]]], [[DST_W]]
-; R600-CHECK-DAG: 16
-; R600-CHECK-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_W_CHAN]]
-; R600-CHECK-DAG: 16
-; SI-CHECK: buffer_load_sshort
-; SI-CHECK: buffer_load_sshort
-; SI-CHECK: buffer_load_sshort
-; SI-CHECK: buffer_load_sshort
+; R600-DAG: VTX_READ_16 [[DST_X:T[0-9]\.[XYZW]]], [[DST_X]]
+; R600-DAG: VTX_READ_16 [[DST_Y:T[0-9]\.[XYZW]]], [[DST_Y]]
+; R600-DAG: VTX_READ_16 [[DST_Z:T[0-9]\.[XYZW]]], [[DST_Z]]
+; R600-DAG: VTX_READ_16 [[DST_W:T[0-9]\.[XYZW]]], [[DST_W]]
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_X_CHAN:[XYZW]]], [[DST_X]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_X_CHAN]]
+; R600-DAG: 16
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Y_CHAN:[XYZW]]], [[DST_Y]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Y_CHAN]]
+; R600-DAG: 16
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_Z_CHAN:[XYZW]]], [[DST_Z]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_Z_CHAN]]
+; R600-DAG: 16
+; R600-DAG: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_W_CHAN:[XYZW]]], [[DST_W]]
+; R600-DAG: 16
+; R600-DAG: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_W_CHAN]]
+; R600-DAG: 16
+; SI: buffer_load_sshort
+; SI: buffer_load_sshort
+; SI: buffer_load_sshort
+; SI: buffer_load_sshort
define void @load_v4i16_sext(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) {
entry:
%0 = load <4 x i16> addrspace(1)* %in
@@ -230,9 +231,9 @@ entry:
; load an i32 value from the global address space.
; FUNC-LABEL: {{^}}load_i32:
-; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+; R600: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
-; SI-CHECK: buffer_load_dword v{{[0-9]+}}
+; SI: buffer_load_dword v{{[0-9]+}}
define void @load_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
entry:
%0 = load i32 addrspace(1)* %in
@@ -242,9 +243,9 @@ entry:
; load a f32 value from the global address space.
; FUNC-LABEL: {{^}}load_f32:
-; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+; R600: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
-; SI-CHECK: buffer_load_dword v{{[0-9]+}}
+; SI: buffer_load_dword v{{[0-9]+}}
define void @load_f32(float addrspace(1)* %out, float addrspace(1)* %in) {
entry:
%0 = load float addrspace(1)* %in
@@ -254,9 +255,9 @@ entry:
; load a v2f32 value from the global address space
; FUNC-LABEL: {{^}}load_v2f32:
-; R600-CHECK: MEM_RAT
-; R600-CHECK: VTX_READ_64
-; SI-CHECK: buffer_load_dwordx2
+; R600: MEM_RAT
+; R600: VTX_READ_64
+; SI: buffer_load_dwordx2
define void @load_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in) {
entry:
%0 = load <2 x float> addrspace(1)* %in
@@ -265,8 +266,8 @@ entry:
}
; FUNC-LABEL: {{^}}load_i64:
-; R600-CHECK: VTX_READ_64
-; SI-CHECK: buffer_load_dwordx2
+; R600: VTX_READ_64
+; SI: buffer_load_dwordx2
define void @load_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
entry:
%0 = load i64 addrspace(1)* %in
@@ -275,11 +276,11 @@ entry:
}
; FUNC-LABEL: {{^}}load_i64_sext:
-; R600-CHECK: MEM_RAT
-; R600-CHECK: MEM_RAT
-; R600-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, T{{[0-9]\.[XYZW]}}, literal.x
-; R600-CHECK: 31
-; SI-CHECK: buffer_load_dword
+; R600: MEM_RAT
+; R600: MEM_RAT
+; R600: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, T{{[0-9]\.[XYZW]}}, literal.x
+; R600: 31
+; SI: buffer_load_dword
define void @load_i64_sext(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
entry:
@@ -290,8 +291,8 @@ entry:
}
; FUNC-LABEL: {{^}}load_i64_zext:
-; R600-CHECK: MEM_RAT
-; R600-CHECK: MEM_RAT
+; R600: MEM_RAT
+; R600: MEM_RAT
define void @load_i64_zext(i64 addrspace(1)* %out, i32 addrspace(1)* %in) {
entry:
%0 = load i32 addrspace(1)* %in
@@ -301,17 +302,17 @@ entry:
}
; FUNC-LABEL: {{^}}load_v8i32:
-; R600-CHECK: VTX_READ_128
-; R600-CHECK: VTX_READ_128
+; R600: VTX_READ_128
+; R600: VTX_READ_128
; XXX: We should be using DWORDX4 instructions on SI.
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
define void @load_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> addrspace(1)* %in) {
entry:
%0 = load <8 x i32> addrspace(1)* %in
@@ -320,27 +321,27 @@ entry:
}
; FUNC-LABEL: {{^}}load_v16i32:
-; R600-CHECK: VTX_READ_128
-; R600-CHECK: VTX_READ_128
-; R600-CHECK: VTX_READ_128
-; R600-CHECK: VTX_READ_128
+; R600: VTX_READ_128
+; R600: VTX_READ_128
+; R600: VTX_READ_128
+; R600: VTX_READ_128
; XXX: We should be using DWORDX4 instructions on SI.
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
-; SI-CHECK: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
+; SI: buffer_load_dword
define void @load_v16i32(<16 x i32> addrspace(1)* %out, <16 x i32> addrspace(1)* %in) {
entry:
%0 = load <16 x i32> addrspace(1)* %in
@@ -354,12 +355,12 @@ entry:
; Load a sign-extended i8 value
; FUNC-LABEL: {{^}}load_const_i8_sext:
-; R600-CHECK: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]]
-; R600-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
-; R600-CHECK: 24
-; R600-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
-; R600-CHECK: 24
-; SI-CHECK: buffer_load_sbyte v{{[0-9]+}},
+; R600: VTX_READ_8 [[DST:T[0-9]\.[XYZW]]], [[DST]]
+; R600: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
+; R600: 24
+; R600: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
+; R600: 24
+; SI: buffer_load_sbyte v{{[0-9]+}},
define void @load_const_i8_sext(i32 addrspace(1)* %out, i8 addrspace(2)* %in) {
entry:
%0 = load i8 addrspace(2)* %in
@@ -370,8 +371,8 @@ entry:
; Load an aligned i8 value
; FUNC-LABEL: {{^}}load_const_i8_aligned:
-; R600-CHECK: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ubyte v{{[0-9]+}},
+; R600: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
+; SI: buffer_load_ubyte v{{[0-9]+}},
define void @load_const_i8_aligned(i32 addrspace(1)* %out, i8 addrspace(2)* %in) {
entry:
%0 = load i8 addrspace(2)* %in
@@ -382,8 +383,8 @@ entry:
; Load an un-aligned i8 value
; FUNC-LABEL: {{^}}load_const_i8_unaligned:
-; R600-CHECK: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ubyte v{{[0-9]+}},
+; R600: VTX_READ_8 T{{[0-9]+\.X, T[0-9]+\.X}}
+; SI: buffer_load_ubyte v{{[0-9]+}},
define void @load_const_i8_unaligned(i32 addrspace(1)* %out, i8 addrspace(2)* %in) {
entry:
%0 = getelementptr i8 addrspace(2)* %in, i32 1
@@ -395,12 +396,12 @@ entry:
; Load a sign-extended i16 value
; FUNC-LABEL: {{^}}load_const_i16_sext:
-; R600-CHECK: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]]
-; R600-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
-; R600-CHECK: 16
-; R600-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
-; R600-CHECK: 16
-; SI-CHECK: buffer_load_sshort
+; R600: VTX_READ_16 [[DST:T[0-9]\.[XYZW]]], [[DST]]
+; R600: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], [[DST]]
+; R600: 16
+; R600: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]]
+; R600: 16
+; SI: buffer_load_sshort
define void @load_const_i16_sext(i32 addrspace(1)* %out, i16 addrspace(2)* %in) {
entry:
%0 = load i16 addrspace(2)* %in
@@ -411,8 +412,8 @@ entry:
; Load an aligned i16 value
; FUNC-LABEL: {{^}}load_const_i16_aligned:
-; R600-CHECK: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ushort
+; R600: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
+; SI: buffer_load_ushort
define void @load_const_i16_aligned(i32 addrspace(1)* %out, i16 addrspace(2)* %in) {
entry:
%0 = load i16 addrspace(2)* %in
@@ -423,8 +424,8 @@ entry:
; Load an un-aligned i16 value
; FUNC-LABEL: {{^}}load_const_i16_unaligned:
-; R600-CHECK: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
-; SI-CHECK: buffer_load_ushort
+; R600: VTX_READ_16 T{{[0-9]+\.X, T[0-9]+\.X}}
+; SI: buffer_load_ushort
define void @load_const_i16_unaligned(i32 addrspace(1)* %out, i16 addrspace(2)* %in) {
entry:
%0 = getelementptr i16 addrspace(2)* %in, i32 1
@@ -436,9 +437,9 @@ entry:
; Load an i32 value from the constant address space.
; FUNC-LABEL: {{^}}load_const_addrspace_i32:
-; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+; R600: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
-; SI-CHECK: s_load_dword s{{[0-9]+}}
+; SI: s_load_dword s{{[0-9]+}}
define void @load_const_addrspace_i32(i32 addrspace(1)* %out, i32 addrspace(2)* %in) {
entry:
%0 = load i32 addrspace(2)* %in
@@ -448,9 +449,9 @@ entry:
; Load a f32 value from the constant address space.
; FUNC-LABEL: {{^}}load_const_addrspace_f32:
-; R600-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+; R600: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
-; SI-CHECK: s_load_dword s{{[0-9]+}}
+; SI: s_load_dword s{{[0-9]+}}
define void @load_const_addrspace_f32(float addrspace(1)* %out, float addrspace(2)* %in) {
%1 = load float addrspace(2)* %in
store float %1, float addrspace(1)* %out
@@ -463,10 +464,10 @@ define void @load_const_addrspace_f32(float addrspace(1)* %out, float addrspace(
; Load an i8 value from the local address space.
; FUNC-LABEL: {{^}}load_i8_local:
-; R600-CHECK: LDS_UBYTE_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u8
+; R600: LDS_UBYTE_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u8
define void @load_i8_local(i32 addrspace(1)* %out, i8 addrspace(3)* %in) {
%1 = load i8 addrspace(3)* %in
%2 = zext i8 %1 to i32
@@ -475,11 +476,11 @@ define void @load_i8_local(i32 addrspace(1)* %out, i8 addrspace(3)* %in) {
}
; FUNC-LABEL: {{^}}load_i8_sext_local:
-; R600-CHECK: LDS_UBYTE_READ_RET
-; R600-CHECK: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i8
+; R600: LDS_UBYTE_READ_RET
+; R600: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i8
define void @load_i8_sext_local(i32 addrspace(1)* %out, i8 addrspace(3)* %in) {
entry:
%0 = load i8 addrspace(3)* %in
@@ -489,12 +490,12 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i8_local:
-; R600-CHECK: LDS_UBYTE_READ_RET
-; R600-CHECK: LDS_UBYTE_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u8
-; SI-CHECK: ds_read_u8
+; R600: LDS_UBYTE_READ_RET
+; R600: LDS_UBYTE_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u8
+; SI: ds_read_u8
define void @load_v2i8_local(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(3)* %in) {
entry:
%0 = load <2 x i8> addrspace(3)* %in
@@ -504,14 +505,14 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i8_sext_local:
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i8
-; SI-CHECK: ds_read_i8
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i8
+; SI: ds_read_i8
define void @load_v2i8_sext_local(<2 x i32> addrspace(1)* %out, <2 x i8> addrspace(3)* %in) {
entry:
%0 = load <2 x i8> addrspace(3)* %in
@@ -521,16 +522,16 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i8_local:
-; R600-CHECK: LDS_UBYTE_READ_RET
-; R600-CHECK: LDS_UBYTE_READ_RET
-; R600-CHECK: LDS_UBYTE_READ_RET
-; R600-CHECK: LDS_UBYTE_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u8
-; SI-CHECK: ds_read_u8
-; SI-CHECK: ds_read_u8
-; SI-CHECK: ds_read_u8
+; R600: LDS_UBYTE_READ_RET
+; R600: LDS_UBYTE_READ_RET
+; R600: LDS_UBYTE_READ_RET
+; R600: LDS_UBYTE_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u8
+; SI: ds_read_u8
+; SI: ds_read_u8
+; SI: ds_read_u8
define void @load_v4i8_local(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(3)* %in) {
entry:
%0 = load <4 x i8> addrspace(3)* %in
@@ -540,20 +541,20 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i8_sext_local:
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: LDS_UBYTE_READ_RET
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i8
-; SI-CHECK: ds_read_i8
-; SI-CHECK: ds_read_i8
-; SI-CHECK: ds_read_i8
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: LDS_UBYTE_READ_RET
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i8
+; SI: ds_read_i8
+; SI: ds_read_i8
+; SI: ds_read_i8
define void @load_v4i8_sext_local(<4 x i32> addrspace(1)* %out, <4 x i8> addrspace(3)* %in) {
entry:
%0 = load <4 x i8> addrspace(3)* %in
@@ -564,10 +565,10 @@ entry:
; Load an i16 value from the local address space.
; FUNC-LABEL: {{^}}load_i16_local:
-; R600-CHECK: LDS_USHORT_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u16
+; R600: LDS_USHORT_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u16
define void @load_i16_local(i32 addrspace(1)* %out, i16 addrspace(3)* %in) {
entry:
%0 = load i16 addrspace(3)* %in
@@ -577,11 +578,11 @@ entry:
}
; FUNC-LABEL: {{^}}load_i16_sext_local:
-; R600-CHECK: LDS_USHORT_READ_RET
-; R600-CHECK: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i16
+; R600: LDS_USHORT_READ_RET
+; R600: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i16
define void @load_i16_sext_local(i32 addrspace(1)* %out, i16 addrspace(3)* %in) {
entry:
%0 = load i16 addrspace(3)* %in
@@ -591,12 +592,12 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i16_local:
-; R600-CHECK: LDS_USHORT_READ_RET
-; R600-CHECK: LDS_USHORT_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u16
-; SI-CHECK: ds_read_u16
+; R600: LDS_USHORT_READ_RET
+; R600: LDS_USHORT_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u16
+; SI: ds_read_u16
define void @load_v2i16_local(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(3)* %in) {
entry:
%0 = load <2 x i16> addrspace(3)* %in
@@ -606,14 +607,14 @@ entry:
}
; FUNC-LABEL: {{^}}load_v2i16_sext_local:
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i16
-; SI-CHECK: ds_read_i16
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i16
+; SI: ds_read_i16
define void @load_v2i16_sext_local(<2 x i32> addrspace(1)* %out, <2 x i16> addrspace(3)* %in) {
entry:
%0 = load <2 x i16> addrspace(3)* %in
@@ -623,16 +624,16 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i16_local:
-; R600-CHECK: LDS_USHORT_READ_RET
-; R600-CHECK: LDS_USHORT_READ_RET
-; R600-CHECK: LDS_USHORT_READ_RET
-; R600-CHECK: LDS_USHORT_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_u16
-; SI-CHECK: ds_read_u16
-; SI-CHECK: ds_read_u16
-; SI-CHECK: ds_read_u16
+; R600: LDS_USHORT_READ_RET
+; R600: LDS_USHORT_READ_RET
+; R600: LDS_USHORT_READ_RET
+; R600: LDS_USHORT_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_u16
+; SI: ds_read_u16
+; SI: ds_read_u16
+; SI: ds_read_u16
define void @load_v4i16_local(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(3)* %in) {
entry:
%0 = load <4 x i16> addrspace(3)* %in
@@ -642,20 +643,20 @@ entry:
}
; FUNC-LABEL: {{^}}load_v4i16_sext_local:
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: LDS_USHORT_READ_RET
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; R600-CHECK-DAG: ASHR
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_i16
-; SI-CHECK: ds_read_i16
-; SI-CHECK: ds_read_i16
-; SI-CHECK: ds_read_i16
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: LDS_USHORT_READ_RET
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; R600-DAG: ASHR
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_i16
+; SI: ds_read_i16
+; SI: ds_read_i16
+; SI: ds_read_i16
define void @load_v4i16_sext_local(<4 x i32> addrspace(1)* %out, <4 x i16> addrspace(3)* %in) {
entry:
%0 = load <4 x i16> addrspace(3)* %in
@@ -666,10 +667,10 @@ entry:
; load an i32 value from the local address space.
; FUNC-LABEL: {{^}}load_i32_local:
-; R600-CHECK: LDS_READ_RET
-; SI-CHECK-NOT: s_wqm_b64
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_b32
+; R600: LDS_READ_RET
+; SI-NOT: s_wqm_b64
+; SI: s_mov_b32 m0
+; SI: ds_read_b32
define void @load_i32_local(i32 addrspace(1)* %out, i32 addrspace(3)* %in) {
entry:
%0 = load i32 addrspace(3)* %in
@@ -679,9 +680,9 @@ entry:
; load a f32 value from the local address space.
; FUNC-LABEL: {{^}}load_f32_local:
-; R600-CHECK: LDS_READ_RET
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_b32
+; R600: LDS_READ_RET
+; SI: s_mov_b32 m0
+; SI: ds_read_b32
define void @load_f32_local(float addrspace(1)* %out, float addrspace(3)* %in) {
entry:
%0 = load float addrspace(3)* %in
@@ -691,10 +692,10 @@ entry:
; load a v2f32 value from the local address space
; FUNC-LABEL: {{^}}load_v2f32_local:
-; R600-CHECK: LDS_READ_RET
-; R600-CHECK: LDS_READ_RET
-; SI-CHECK: s_mov_b32 m0
-; SI-CHECK: ds_read_b64
+; R600: LDS_READ_RET
+; R600: LDS_READ_RET
+; SI: s_mov_b32 m0
+; SI: ds_read_b64
define void @load_v2f32_local(<2 x float> addrspace(1)* %out, <2 x float> addrspace(3)* %in) {
entry:
%0 = load <2 x float> addrspace(3)* %in
@@ -704,11 +705,11 @@ entry:
; Test loading a i32 and v2i32 value from the same base pointer.
; FUNC-LABEL: {{^}}load_i32_v2i32_local:
-; R600-CHECK: LDS_READ_RET
-; R600-CHECK: LDS_READ_RET
-; R600-CHECK: LDS_READ_RET
-; SI-CHECK-DAG: ds_read_b32
-; SI-CHECK-DAG: ds_read2_b32
+; R600: LDS_READ_RET
+; R600: LDS_READ_RET
+; R600: LDS_READ_RET
+; SI-DAG: ds_read_b32
+; SI-DAG: ds_read2_b32
define void @load_i32_v2i32_local(<2 x i32> addrspace(1)* %out, i32 addrspace(3)* %in) {
%scalar = load i32 addrspace(3)* %in
%tmp0 = bitcast i32 addrspace(3)* %in to <2 x i32> addrspace(3)*
@@ -726,9 +727,9 @@ define void @load_i32_v2i32_local(<2 x i32> addrspace(1)* %out, i32 addrspace(3)
; On SI we need to make sure that the base offset is a register and not
; an immediate.
; FUNC-LABEL: {{^}}load_i32_local_const_ptr:
-; SI-CHECK: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0
-; SI-CHECK: ds_read_b32 v0, v[[ZERO]] offset:4
-; R600-CHECK: LDS_READ_RET
+; SI: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0
+; SI: ds_read_b32 v0, v[[ZERO]] offset:4
+; R600: LDS_READ_RET
define void @load_i32_local_const_ptr(i32 addrspace(1)* %out, i32 addrspace(3)* %in) {
entry:
%tmp0 = getelementptr [512 x i32] addrspace(3)* @lds, i32 0, i32 1
generated by cgit v1.1 at 2024-07-23 13:37:52 +0000