[flang-commits] [flang] ae1c064 - [flang] Adding two more PowerPC MMA intrinsics
Daniel Chen via flang-commits
flang-commits at lists.llvm.org
Mon Aug 21 08:00:41 PDT 2023
Author: Daniel Chen
Date: 2023-08-21T11:00:19-04:00
New Revision: ae1c0645aa7c11f58641d80293f217990b94b328
URL: https://github.com/llvm/llvm-project/commit/ae1c0645aa7c11f58641d80293f217990b94b328
DIFF: https://github.com/llvm/llvm-project/commit/ae1c0645aa7c11f58641d80293f217990b94b328.diff
LOG: [flang] Adding two more PowerPC MMA intrinsics
Adding mma_lxvp and mma_stxvp, which are alias to already landed vec_lxvp and vec_stxvp respectively.
Differential Revision: https://reviews.llvm.org/D158274
Added:
Modified:
flang/module/mma.f90
flang/test/Lower/PowerPC/ppc-pwr10-vec-intrinsics.f90
Removed:
################################################################################
diff --git a/flang/module/mma.f90 b/flang/module/mma.f90
index d6d2eb87879bc0..4c41822e000a3d 100644
--- a/flang/module/mma.f90
+++ b/flang/module/mma.f90
@@ -18,7 +18,111 @@ elemental subroutine sub_vq(acc)
__vector_quad, intent(inout) :: acc
end subroutine
+!! ========== 2 argument subroutine interface ================================!!
+!! __vector_pair function f(i, vector(i))
+#define FUNC_VPI0VI(VKIND) \
+ pure __vector_pair function func_vpi0vi##VKIND(arg1, arg2); \
+ integer(8), intent(in) :: arg1; \
+ !dir$ ignore_tkr(k) arg1; \
+ vector(integer(VKIND)), intent(in) :: arg2; \
+ !dir$ ignore_tkr(r) arg2; \
+ end function;
+
+! __vector_pair function f(i, vector(u))
+#define FUNC_VPI0VU(VKIND) \
+ pure __vector_pair function func_vpi0vu##VKIND(arg1, arg2); \
+ integer(8), intent(in) :: arg1; \
+ !dir$ ignore_tkr(k) arg1; \
+ vector(unsigned(VKIND)), intent(in) :: arg2; \
+ !dir$ ignore_tkr(r) arg2; \
+ end function;
+
+! __vector_pair function f(i, vector(r))
+#define FUNC_VPI0VR(VKIND) \
+ pure __vector_pair function func_vpi0vr##VKIND(arg1, arg2); \
+ integer(8), intent(in) :: arg1; \
+ !dir$ ignore_tkr(k) arg1; \
+ vector(real(VKIND)), intent(in) :: arg2; \
+ !dir$ ignore_tkr(r) arg2; \
+ end function;
+
+! __vector_pair function f(i, __vector_pair)
+#define FUNC_VPI0VP \
+ pure __vector_pair function func_vpi0vp(arg1, arg2); \
+ integer(8), intent(in) :: arg1; \
+ !dir$ ignore_tkr(k) arg1; \
+ __vector_pair, intent(in) :: arg2; \
+ !dir$ ignore_tkr(r) arg2; \
+ end function;
+
+ FUNC_VPI0VI(1) FUNC_VPI0VI(2) FUNC_VPI0VI(4) FUNC_VPI0VI(8)
+ FUNC_VPI0VU(1) FUNC_VPI0VU(2) FUNC_VPI0VU(4) FUNC_VPI0VU(8)
+ FUNC_VPI0VR(4) FUNC_VPI0VR(8)
+ FUNC_VPI0VP
+
+#undef FUNC_VPI0VP
+#undef FUNC_VPI0VR
+#undef FUNC_VPI0VU
+#undef FUNC_VPI0VI
+
!! ========== 3 arguments subroutine interface ===============================!!
+!! __vector_pair subroutine s(vp, integer, vector(i))
+#define SUB_VPI0VI(VKIND) \
+ pure subroutine sub_vpi0vi##VKIND(arg1, arg2, arg3); \
+ __vector_pair, intent(in) :: arg1; \
+ integer(8), intent(in) :: arg2; \
+ !dir$ ignore_tkr(k) arg2; \
+ vector(integer(VKIND)), intent(out) :: arg3; \
+ !dir$ ignore_tkr(r) arg3; \
+ end subroutine;
+
+! subroutine(__vector_pair, i, vector(u))
+#define SUB_VPI0VU(VKIND) \
+ pure subroutine sub_vpi0vu##VKIND(arg1, arg2, arg3); \
+ __vector_pair, intent(in) :: arg1; \
+ integer(8), intent(in) :: arg2; \
+ !dir$ ignore_tkr(k) arg2; \
+ vector(unsigned(VKIND)), intent(out) :: arg3; \
+ !dir$ ignore_tkr(r) arg3; \
+ end subroutine;
+
+! subroutine(__vector_pair, i, vector(r))
+#define SUB_VPI0VR(VKIND) \
+ pure subroutine sub_vpi0vr##VKIND(arg1, arg2, arg3); \
+ __vector_pair, intent(in) :: arg1; \
+ integer(8), intent(in) :: arg2; \
+ !dir$ ignore_tkr(k) arg2; \
+ vector(real(VKIND)), intent(out) :: arg3; \
+ !dir$ ignore_tkr(r) arg3; \
+ end subroutine;
+
+! subroutine(__vector_pair, i, i)
+ pure subroutine sub_vpi0i0(arg1, arg2, arg3)
+ __vector_pair, intent(in) :: arg1
+ integer(8), intent(in) :: arg2
+ !dir$ ignore_tkr(k) arg2
+ integer(8), intent(out) :: arg3
+ !dir$ ignore_tkr(kr) arg3
+ end subroutine
+
+! subroutine(__vector_pair, i, r)
+ pure subroutine sub_vpi0r0(arg1, arg2, arg3)
+ __vector_pair, intent(in) :: arg1
+ integer(8), intent(in) :: arg2
+ !dir$ ignore_tkr(k) arg2
+ real(8), intent(out) :: arg3
+ !dir$ ignore_tkr(kr) arg3
+ end subroutine
+
+! subroutine(__vector_pair, i, __vector_pair)
+ pure subroutine sub_vpi0vp(arg1, arg2, arg3)
+ __vector_pair, intent(in) :: arg1
+ integer(8), intent(in) :: arg2
+ !dir$ ignore_tkr(k) arg2
+ __vector_pair, intent(out) :: arg3
+ !dir$ ignore_tkr(r) arg3
+ end subroutine
+
!! subroutine s(__vector_pair, vector(i), vector(i))
#define ELEM_SUB_VPVIVI(VKIND) \
elemental subroutine sub_vpvi##VKIND##vi##VKIND(pair, arg1, arg2); \
@@ -45,10 +149,16 @@ elemental subroutine sub_vpvr##VKIND##vr##VKIND(pair, arg1, arg2); \
ELEM_SUB_VPVUVU(1) ELEM_SUB_VPVUVU(2)
ELEM_SUB_VPVUVU(4) ELEM_SUB_VPVUVU(8)
ELEM_SUB_VPVRVR(4) ELEM_SUB_VPVRVR(8)
+ SUB_VPI0VI(1) SUB_VPI0VI(2) SUB_VPI0VI(4) SUB_VPI0VI(8)
+ SUB_VPI0VU(1) SUB_VPI0VU(2) SUB_VPI0VU(4) SUB_VPI0VU(8)
+ SUB_VPI0VR(4) SUB_VPI0VR(8)
#undef ELEM_SUB_VPVIVI
#undef ELEM_SUB_VPVUVU
#undef ELEM_SUB_VPVRVR
+#undef SUB_VPI0VR
+#undef SUB_VPI0VU
+#undef SUB_VPI0VI
!! subroutine s(__vector_quad, vector(i), vector(i))
#define ELEM_SUB_VQVIVI(INTENT, VKIND) \
@@ -246,6 +356,81 @@ elemental subroutine sub_atvq(data, acc)
end interface
+!-------------------------------------------------------
+! __vector_pair function(integer, vector/__vector_pair)
+!-------------------------------------------------------
+#define VP_I0_VI(NAME, VKIND) __ppc_##NAME##_vpi0##vi##VKIND
+#define VP_I0_VU(NAME, VKIND) __ppc_##NAME##_vpi0##vu##VKIND
+#define VP_I0_VR(NAME, VKIND) __ppc_##NAME##_vpi0##vr##VKIND
+#define VP_I0_VP(NAME) __ppc_##NAME##_vpi0vp0
+
+#define VEC_VP_I0_VI(NAME, VKIND) \
+ procedure(func_vpi0vi##VKIND) :: VP_I0_VI(NAME, VKIND);
+#define VEC_VP_I0_VU(NAME, VKIND) \
+ procedure(func_vpi0vu##VKIND) :: VP_I0_VU(NAME, VKIND);
+#define VEC_VP_I0_VR(NAME, VKIND) \
+ procedure(func_vpi0vr##VKIND) :: VP_I0_VR(NAME, VKIND);
+#define VEC_VP_I0_VP(NAME) procedure(func_vpi0vp) :: VP_I0_VP(NAME);
+
+! mma_lxvp (using vec_lxvp)
+ VEC_VP_I0_VI(vec_lxvp,1) VEC_VP_I0_VI(vec_lxvp,2) VEC_VP_I0_VI(vec_lxvp,4) VEC_VP_I0_VI(vec_lxvp,8)
+ VEC_VP_I0_VU(vec_lxvp,1) VEC_VP_I0_VU(vec_lxvp,2) VEC_VP_I0_VU(vec_lxvp,4) VEC_VP_I0_VU(vec_lxvp,8)
+ VEC_VP_I0_VR(vec_lxvp,4) VEC_VP_I0_VR(vec_lxvp,8)
+ VEC_VP_I0_VP(vec_lxvp)
+ interface mma_lxvp
+ procedure :: VP_I0_VI(vec_lxvp,1), VP_I0_VI(vec_lxvp,2), VP_I0_VI(vec_lxvp,4), VP_I0_VI(vec_lxvp,8)
+ procedure :: VP_I0_VU(vec_lxvp,1), VP_I0_VU(vec_lxvp,2), VP_I0_VU(vec_lxvp,4), VP_I0_VU(vec_lxvp,8)
+ procedure :: VP_I0_VR(vec_lxvp,4), VP_I0_VR(vec_lxvp,8)
+ procedure :: VP_I0_VP(vec_lxvp)
+ end interface mma_lxvp
+ public :: mma_lxvp
+
+#undef VEC_VP_I0_VP
+#undef VEC_VP_I0_VR
+#undef VEC_VP_I0_VU
+#undef VEC_VP_I0_VI
+#undef VP_I0_VP
+#undef VP_I0_VR
+#undef VP_I0_VU
+#undef VP_I0_VI
+
+!-----------------------------------------------------------------------
+! subroutine(__vector_pair, integer, __vector_pair/vector/integer/real)
+!-----------------------------------------------------------------------
+#define VP_I0_VI(NAME, VKIND) __ppc_##NAME##_vpi0vi##VKIND
+#define VP_I0_VU(NAME, VKIND) __ppc_##NAME##_vpi0vu##VKIND
+#define VP_I0_VR(NAME, VKIND) __ppc_##NAME##_vpi0vr##VKIND
+
+#define VEC_VP_I0_VI(NAME, VKIND) \
+ procedure(sub_vpi0vi##VKIND) :: VP_I0_VI(NAME, VKIND);
+#define VEC_VP_I0_VU(NAME, VKIND) \
+ procedure(sub_vpi0vu##VKIND) :: VP_I0_VU(NAME, VKIND);
+#define VEC_VP_I0_VR(NAME, VKIND) \
+ procedure(sub_vpi0vr##VKIND) :: VP_I0_VR(NAME, VKIND);
+
+! mma_stxvp (using vec_stxvp)
+ procedure(sub_vpi0vp) :: __ppc_vec_stxvp_vpi0vp0
+ procedure(sub_vpi0i0) :: __ppc_vec_stxvp_vpi0i0
+ procedure(sub_vpi0r0) :: __ppc_vec_stxvp_vpi0r0
+ VEC_VP_I0_VI(vec_stxvp, 1) VEC_VP_I0_VI(vec_stxvp, 2) VEC_VP_I0_VI(vec_stxvp, 4) VEC_VP_I0_VI(vec_stxvp, 8)
+ VEC_VP_I0_VU(vec_stxvp, 1) VEC_VP_I0_VU(vec_stxvp, 2) VEC_VP_I0_VU(vec_stxvp, 4) VEC_VP_I0_VU(vec_stxvp, 8)
+ VEC_VP_I0_VR(vec_stxvp, 4) VEC_VP_I0_VR(vec_stxvp, 8)
+ interface mma_stxvp
+ procedure :: __ppc_vec_stxvp_vpi0vp0
+ procedure :: __ppc_vec_stxvp_vpi0i0
+ procedure :: __ppc_vec_stxvp_vpi0r0
+ procedure :: VP_I0_VI(vec_stxvp, 1), VP_I0_VI(vec_stxvp, 2), VP_I0_VI(vec_stxvp, 4), VP_I0_VI(vec_stxvp, 8)
+ procedure :: VP_I0_VU(vec_stxvp, 1), VP_I0_VU(vec_stxvp, 2), VP_I0_VU(vec_stxvp, 4), VP_I0_VU(vec_stxvp, 8)
+ procedure :: VP_I0_VR(vec_stxvp, 4), VP_I0_VR(vec_stxvp, 8)
+ end interface mma_stxvp
+ public :: mma_stxvp
+
+#undef VEC_VP_I0_VR
+#undef VEC_VP_I0_VU
+#undef VEC_VP_I0_VI
+#undef VP_I0_VR
+#undef VP_I0_VU
+#undef VP_I0_VI
#define SUB_VQ_VI_VI_VI_VI(NAME, VKIND) __ppc_##NAME##_vi##VKIND##vi##VKIND##vi##VKIND##vi##VKIND
#define SUB_VQ_VU_VU_VU_VU(NAME, VKIND) __ppc_##NAME##_vu##VKIND##vu##VKIND##vu##VKIND##vu##VKIND
diff --git a/flang/test/Lower/PowerPC/ppc-pwr10-vec-intrinsics.f90 b/flang/test/Lower/PowerPC/ppc-pwr10-vec-intrinsics.f90
index 92875567503ba3..07e8795b4bc783 100644
--- a/flang/test/Lower/PowerPC/ppc-pwr10-vec-intrinsics.f90
+++ b/flang/test/Lower/PowerPC/ppc-pwr10-vec-intrinsics.f90
@@ -1,5 +1,24 @@
! RUN: %flang_fc1 -triple powerpc64le-unknown-unknown -target-cpu pwr10 -emit-llvm %s -o - | FileCheck --check-prefixes="CHECK" %s
! REQUIRES: target=powerpc{{.*}}
+
+!----------------------
+! mma_lxvp
+!----------------------
+
+ subroutine mma_lxvp_test_i2(v1, offset, vp)
+ use, intrinsic :: mma
+ integer(2) :: offset
+ vector(integer(2)) :: v1
+ __vector_pair :: vp
+ vp = mma_lxvp(offset, v1)
+ end subroutine mma_lxvp_test_i2
+
+!CHECK-LABEL: @mma_lxvp_test_i2_
+!CHECK: %[[offset:.*]] = load i16, ptr %1, align 2
+!CHECK: %[[addr:.*]] = getelementptr i8, ptr %0, i16 %[[offset]]
+!CHECK: %[[call:.*]] = call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %[[addr]])
+!CHECK: store <256 x i1> %[[call]], ptr %2, align 32
+
subroutine test_cvspbf16()
implicit none
vector(unsigned(1)) :: v1, v2
@@ -268,6 +287,24 @@ end subroutine vsx_lxvp_test_vp_arr
!CHECK: %[[call:.*]] = call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %[[addr]])
!CHECK: store <256 x i1> %[[call]], ptr %2, align 32
+!----------------------
+! mma_stxvp
+!----------------------
+
+ subroutine test_mma_stxvp_i1(vp, offset, v1)
+ use, intrinsic :: mma
+ integer(1) :: offset
+ vector(integer(2)) :: v1
+ __vector_pair :: vp
+ call mma_stxvp(vp, offset, v1)
+ end subroutine test_mma_stxvp_i1
+
+!CHECK-LABEL: @test_mma_stxvp_i1_
+!CHECK: %[[vp:.*]] = load <256 x i1>, ptr %0, align 32
+!CHECK: %[[offset:.*]] = load i8, ptr %1, align 1
+!CHECK: %[[addr:.*]] = getelementptr i8, ptr %2, i8 %[[offset]]
+!CHECK: call void @llvm.ppc.vsx.stxvp(<256 x i1> %[[vp]], ptr %[[addr]])
+
!----------------------
! vec_stxvp
!----------------------
More information about the flang-commits
mailing list