; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
; RUN: opt < %s -mtriple=nvptx64-nvidia-cuda -S -passes=separate-const-offset-from-gep,gvn \
; RUN: -reassociate-geps-verify-no-dead-code \
; RUN: | FileCheck %s --check-prefix=IR
; RUN: llc < %s -mtriple=nvptx64-nvidia-cuda -mcpu=sm_20 \
; RUN: | FileCheck %s --check-prefix=PTX
; Verifies the SeparateConstOffsetFromGEP pass.
; The following code computes
; *output = array[x][y] + array[x][y+1] + array[x+1][y] + array[x+1][y+1]
;
; We expect SeparateConstOffsetFromGEP to transform it to
;
; ptr base = &a[x][y];
; *output = base[0] + base[1] + base[32] + base[33];
;
; so the backend can emit PTX that uses fewer virtual registers.
@array = internal addrspace(3) global [32 x [32 x float]] zeroinitializer, align 4
define void @sum_of_array(i32 %x, i32 %y, ptr nocapture %output) {
; IR-LABEL: define void @sum_of_array(
; IR-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr nocapture [[OUTPUT:%.*]]) {
; IR-NEXT: .preheader:
; IR-NEXT: [[TMP0:%.*]] = sext i32 [[Y]] to i64
; IR-NEXT: [[TMP1:%.*]] = sext i32 [[X]] to i64
; IR-NEXT: [[TMP2:%.*]] = getelementptr [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 [[TMP1]], i64 [[TMP0]]
; IR-NEXT: [[TMP3:%.*]] = addrspacecast ptr addrspace(3) [[TMP2]] to ptr
; IR-NEXT: [[TMP4:%.*]] = load float, ptr [[TMP3]], align 4
; IR-NEXT: [[TMP5:%.*]] = fadd float [[TMP4]], 0.000000e+00
; IR-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 4
; IR-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(3) [[TMP6]] to ptr
; IR-NEXT: [[TMP8:%.*]] = load float, ptr [[TMP7]], align 4
; IR-NEXT: [[TMP9:%.*]] = fadd float [[TMP5]], [[TMP8]]
; IR-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 128
; IR-NEXT: [[TMP11:%.*]] = addrspacecast ptr addrspace(3) [[TMP10]] to ptr
; IR-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP11]], align 4
; IR-NEXT: [[TMP13:%.*]] = fadd float [[TMP9]], [[TMP12]]
; IR-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 132
; IR-NEXT: [[TMP15:%.*]] = addrspacecast ptr addrspace(3) [[TMP14]] to ptr
; IR-NEXT: [[TMP16:%.*]] = load float, ptr [[TMP15]], align 4
; IR-NEXT: [[TMP17:%.*]] = fadd float [[TMP13]], [[TMP16]]
; IR-NEXT: store float [[TMP17]], ptr [[OUTPUT]], align 4
; IR-NEXT: ret void
;
.preheader:
%0 = sext i32 %y to i64
%1 = sext i32 %x to i64
%2 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %0
%3 = addrspacecast ptr addrspace(3) %2 to ptr
%4 = load float, ptr %3, align 4
%5 = fadd float %4, 0.000000e+00
%6 = add i32 %y, 1
%7 = sext i32 %6 to i64
%8 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %7
%9 = addrspacecast ptr addrspace(3) %8 to ptr
%10 = load float, ptr %9, align 4
%11 = fadd float %5, %10
%12 = add i32 %x, 1
%13 = sext i32 %12 to i64
%14 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %13, i64 %0
%15 = addrspacecast ptr addrspace(3) %14 to ptr
%16 = load float, ptr %15, align 4
%17 = fadd float %11, %16
%18 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %13, i64 %7
%19 = addrspacecast ptr addrspace(3) %18 to ptr
%20 = load float, ptr %19, align 4
%21 = fadd float %17, %20
store float %21, ptr %output, align 4
ret void
}
; PTX-LABEL: sum_of_array(
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG:%(rd|r)[0-9]+]]]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+4]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+128]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+132]
; TODO: GVN is unable to preserve the "inbounds" keyword on the first GEP. Need
; some infrastructure changes to enable such optimizations.
; @sum_of_array2 is very similar to @sum_of_array. The only difference is in
; the order of "sext" and "add" when computing the array indices. @sum_of_array
; computes add before sext, e.g., array[sext(x + 1)][sext(y + 1)], while
; @sum_of_array2 computes sext before add,
; e.g., array[sext(x) + 1][sext(y) + 1]. SeparateConstOffsetFromGEP should be
; able to extract constant offsets from both forms.
define void @sum_of_array2(i32 %x, i32 %y, ptr nocapture %output) {
; IR-LABEL: define void @sum_of_array2(
; IR-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr nocapture [[OUTPUT:%.*]]) {
; IR-NEXT: .preheader:
; IR-NEXT: [[TMP0:%.*]] = sext i32 [[Y]] to i64
; IR-NEXT: [[TMP1:%.*]] = sext i32 [[X]] to i64
; IR-NEXT: [[TMP2:%.*]] = getelementptr [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 [[TMP1]], i64 [[TMP0]]
; IR-NEXT: [[TMP3:%.*]] = addrspacecast ptr addrspace(3) [[TMP2]] to ptr
; IR-NEXT: [[TMP4:%.*]] = load float, ptr [[TMP3]], align 4
; IR-NEXT: [[TMP5:%.*]] = fadd float [[TMP4]], 0.000000e+00
; IR-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 4
; IR-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(3) [[TMP6]] to ptr
; IR-NEXT: [[TMP8:%.*]] = load float, ptr [[TMP7]], align 4
; IR-NEXT: [[TMP9:%.*]] = fadd float [[TMP5]], [[TMP8]]
; IR-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 128
; IR-NEXT: [[TMP11:%.*]] = addrspacecast ptr addrspace(3) [[TMP10]] to ptr
; IR-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP11]], align 4
; IR-NEXT: [[TMP13:%.*]] = fadd float [[TMP9]], [[TMP12]]
; IR-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 132
; IR-NEXT: [[TMP15:%.*]] = addrspacecast ptr addrspace(3) [[TMP14]] to ptr
; IR-NEXT: [[TMP16:%.*]] = load float, ptr [[TMP15]], align 4
; IR-NEXT: [[TMP17:%.*]] = fadd float [[TMP13]], [[TMP16]]
; IR-NEXT: store float [[TMP17]], ptr [[OUTPUT]], align 4
; IR-NEXT: ret void
;
.preheader:
%0 = sext i32 %y to i64
%1 = sext i32 %x to i64
%2 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %0
%3 = addrspacecast ptr addrspace(3) %2 to ptr
%4 = load float, ptr %3, align 4
%5 = fadd float %4, 0.000000e+00
%6 = add i64 %0, 1
%7 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %6
%8 = addrspacecast ptr addrspace(3) %7 to ptr
%9 = load float, ptr %8, align 4
%10 = fadd float %5, %9
%11 = add i64 %1, 1
%12 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %11, i64 %0
%13 = addrspacecast ptr addrspace(3) %12 to ptr
%14 = load float, ptr %13, align 4
%15 = fadd float %10, %14
%16 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %11, i64 %6
%17 = addrspacecast ptr addrspace(3) %16 to ptr
%18 = load float, ptr %17, align 4
%19 = fadd float %15, %18
store float %19, ptr %output, align 4
ret void
}
; PTX-LABEL: sum_of_array2(
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG:%(rd|r)[0-9]+]]]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+4]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+128]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+132]
; This function loads
; array[zext(x)][zext(y)]
; array[zext(x)][zext(y +nuw 1)]
; array[zext(x +nuw 1)][zext(y)]
; array[zext(x +nuw 1)][zext(y +nuw 1)].
;
; This function is similar to @sum_of_array, but it
; 1) extends array indices using zext instead of sext;
; 2) annotates the addition with "nuw"; otherwise, zext(x + 1) => zext(x) + 1
; may be invalid.
define void @sum_of_array3(i32 %x, i32 %y, ptr nocapture %output) {
; IR-LABEL: define void @sum_of_array3(
; IR-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr nocapture [[OUTPUT:%.*]]) {
; IR-NEXT: .preheader:
; IR-NEXT: [[TMP0:%.*]] = zext i32 [[Y]] to i64
; IR-NEXT: [[TMP1:%.*]] = zext i32 [[X]] to i64
; IR-NEXT: [[TMP2:%.*]] = getelementptr [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 [[TMP1]], i64 [[TMP0]]
; IR-NEXT: [[TMP3:%.*]] = addrspacecast ptr addrspace(3) [[TMP2]] to ptr
; IR-NEXT: [[TMP4:%.*]] = load float, ptr [[TMP3]], align 4
; IR-NEXT: [[TMP5:%.*]] = fadd float [[TMP4]], 0.000000e+00
; IR-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 4
; IR-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(3) [[TMP6]] to ptr
; IR-NEXT: [[TMP8:%.*]] = load float, ptr [[TMP7]], align 4
; IR-NEXT: [[TMP9:%.*]] = fadd float [[TMP5]], [[TMP8]]
; IR-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 128
; IR-NEXT: [[TMP11:%.*]] = addrspacecast ptr addrspace(3) [[TMP10]] to ptr
; IR-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP11]], align 4
; IR-NEXT: [[TMP13:%.*]] = fadd float [[TMP9]], [[TMP12]]
; IR-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 132
; IR-NEXT: [[TMP15:%.*]] = addrspacecast ptr addrspace(3) [[TMP14]] to ptr
; IR-NEXT: [[TMP16:%.*]] = load float, ptr [[TMP15]], align 4
; IR-NEXT: [[TMP17:%.*]] = fadd float [[TMP13]], [[TMP16]]
; IR-NEXT: store float [[TMP17]], ptr [[OUTPUT]], align 4
; IR-NEXT: ret void
;
.preheader:
%0 = zext i32 %y to i64
%1 = zext i32 %x to i64
%2 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %0
%3 = addrspacecast ptr addrspace(3) %2 to ptr
%4 = load float, ptr %3, align 4
%5 = fadd float %4, 0.000000e+00
%6 = add nuw i32 %y, 1
%7 = zext i32 %6 to i64
%8 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %7
%9 = addrspacecast ptr addrspace(3) %8 to ptr
%10 = load float, ptr %9, align 4
%11 = fadd float %5, %10
%12 = add nuw i32 %x, 1
%13 = zext i32 %12 to i64
%14 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %13, i64 %0
%15 = addrspacecast ptr addrspace(3) %14 to ptr
%16 = load float, ptr %15, align 4
%17 = fadd float %11, %16
%18 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %13, i64 %7
%19 = addrspacecast ptr addrspace(3) %18 to ptr
%20 = load float, ptr %19, align 4
%21 = fadd float %17, %20
store float %21, ptr %output, align 4
ret void
}
; PTX-LABEL: sum_of_array3(
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG:%(rd|r)[0-9]+]]]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+4]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+128]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+132]
; This function loads
; array[zext(x)][zext(y)]
; array[zext(x)][zext(y)]
; array[zext(x) + 1][zext(y) + 1]
; array[zext(x) + 1][zext(y) + 1].
;
; We expect the generated code to reuse the computation of
; &array[zext(x)][zext(y)]. See the expected IR and PTX for details.
define void @sum_of_array4(i32 %x, i32 %y, ptr nocapture %output) {
; IR-LABEL: define void @sum_of_array4(
; IR-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr nocapture [[OUTPUT:%.*]]) {
; IR-NEXT: .preheader:
; IR-NEXT: [[TMP0:%.*]] = zext i32 [[Y]] to i64
; IR-NEXT: [[TMP1:%.*]] = zext i32 [[X]] to i64
; IR-NEXT: [[TMP2:%.*]] = getelementptr [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 [[TMP1]], i64 [[TMP0]]
; IR-NEXT: [[TMP3:%.*]] = addrspacecast ptr addrspace(3) [[TMP2]] to ptr
; IR-NEXT: [[TMP4:%.*]] = load float, ptr [[TMP3]], align 4
; IR-NEXT: [[TMP5:%.*]] = fadd float [[TMP4]], 0.000000e+00
; IR-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 4
; IR-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(3) [[TMP6]] to ptr
; IR-NEXT: [[TMP8:%.*]] = load float, ptr [[TMP7]], align 4
; IR-NEXT: [[TMP9:%.*]] = fadd float [[TMP5]], [[TMP8]]
; IR-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 128
; IR-NEXT: [[TMP11:%.*]] = addrspacecast ptr addrspace(3) [[TMP10]] to ptr
; IR-NEXT: [[TMP12:%.*]] = load float, ptr [[TMP11]], align 4
; IR-NEXT: [[TMP13:%.*]] = fadd float [[TMP9]], [[TMP12]]
; IR-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[TMP2]], i64 132
; IR-NEXT: [[TMP15:%.*]] = addrspacecast ptr addrspace(3) [[TMP14]] to ptr
; IR-NEXT: [[TMP16:%.*]] = load float, ptr [[TMP15]], align 4
; IR-NEXT: [[TMP17:%.*]] = fadd float [[TMP13]], [[TMP16]]
; IR-NEXT: store float [[TMP17]], ptr [[OUTPUT]], align 4
; IR-NEXT: ret void
;
.preheader:
%0 = zext i32 %y to i64
%1 = zext i32 %x to i64
%2 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %0
%3 = addrspacecast ptr addrspace(3) %2 to ptr
%4 = load float, ptr %3, align 4
%5 = fadd float %4, 0.000000e+00
%6 = add i64 %0, 1
%7 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %1, i64 %6
%8 = addrspacecast ptr addrspace(3) %7 to ptr
%9 = load float, ptr %8, align 4
%10 = fadd float %5, %9
%11 = add i64 %1, 1
%12 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %11, i64 %0
%13 = addrspacecast ptr addrspace(3) %12 to ptr
%14 = load float, ptr %13, align 4
%15 = fadd float %10, %14
%16 = getelementptr inbounds [32 x [32 x float]], ptr addrspace(3) @array, i64 0, i64 %11, i64 %6
%17 = addrspacecast ptr addrspace(3) %16 to ptr
%18 = load float, ptr %17, align 4
%19 = fadd float %15, %18
store float %19, ptr %output, align 4
ret void
}
; PTX-LABEL: sum_of_array4(
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG:%(rd|r)[0-9]+]]]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+4]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+128]
; PTX-DAG: ld.shared.f32 {{%f[0-9]+}}, [[[BASE_REG]]+132]
; The source code is:
; p0 = &input[sext(x + y)];
; p1 = &input[sext(x + (y + 5))];
;
; Without reuniting extensions, SeparateConstOffsetFromGEP would emit
; p0 = &input[sext(x + y)];
; t1 = &input[sext(x) + sext(y)];
; p1 = &t1[5];
;
; With reuniting extensions, it merges p0 and t1 and thus emits
; p0 = &input[sext(x + y)];
; p1 = &p0[5];
define void @reunion(i32 %x, i32 %y, ptr %input) {
; IR-LABEL: define void @reunion(
; IR-SAME: i32 [[X:%.*]], i32 [[Y:%.*]], ptr [[INPUT:%.*]]) {
; IR-NEXT: entry:
; IR-NEXT: [[XY:%.*]] = add nsw i32 [[X]], [[Y]]
; IR-NEXT: [[TMP0:%.*]] = sext i32 [[XY]] to i64
; IR-NEXT: [[P0:%.*]] = getelementptr float, ptr [[INPUT]], i64 [[TMP0]]
; IR-NEXT: [[V0:%.*]] = load float, ptr [[P0]], align 4
; IR-NEXT: call void @use(float [[V0]])
; IR-NEXT: [[P13:%.*]] = getelementptr inbounds i8, ptr [[P0]], i64 20
; IR-NEXT: [[V1:%.*]] = load float, ptr [[P13]], align 4
; IR-NEXT: call void @use(float [[V1]])
; IR-NEXT: ret void
;
; PTX-LABEL: reunion(
entry:
%xy = add nsw i32 %x, %y
%0 = sext i32 %xy to i64
%p0 = getelementptr inbounds float, ptr %input, i64 %0
%v0 = load float, ptr %p0, align 4
; PTX: ld.f32 %f{{[0-9]+}}, [[[p0:%rd[0-9]+]]]
call void @use(float %v0)
%y5 = add nsw i32 %y, 5
%xy5 = add nsw i32 %x, %y5
%1 = sext i32 %xy5 to i64
%p1 = getelementptr inbounds float, ptr %input, i64 %1
%v1 = load float, ptr %p1, align 4
; PTX: ld.f32 %f{{[0-9]+}}, [[[p0]]+20]
call void @use(float %v1)
ret void
}
declare void @use(float)
Codebase Browser
llvm