llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT--> @llvm/pr-subscribers-clang Author: None (amilendra) <details> <summary>Changes</summary> Add mfloat8_t support for the following Neon load intrinsics. - VLD1 - VLD1_X2 - VLD1_X3 - VLD1_X4 - VLD1_LANE - VLD1_DUP - VLD2 - VLD3 - VLD4 - VLD2_DUP - VLD3_DUP - VLD4_DUP - VLD2_LANE - VLD3_LANE - VLD4_LANE --- Patch is 71.51 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/145666.diff 2 Files Affected: - (modified) clang/include/clang/Basic/arm_neon.td (+15-15) - (added) clang/test/CodeGen/AArch64/neon-fp8-intrinsics/acle_neon_loads.c (+825) ``````````diff diff --git a/clang/include/clang/Basic/arm_neon.td b/clang/include/clang/Basic/arm_neon.td index 7251cc2d1759a..a21773bd315cd 100644 --- a/clang/include/clang/Basic/arm_neon.td +++ b/clang/include/clang/Basic/arm_neon.td @@ -453,18 +453,18 @@ def VSLI_N : WInst<"vsli_n", "...I", //////////////////////////////////////////////////////////////////////////////// // E.3.14 Loads and stores of a single vector def VLD1 : WInst<"vld1", ".(c*!)", - "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">; + "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPsmQm">; def VLD1_X2 : WInst<"vld1_x2", "2(c*!)", - "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">; + "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPsmQm">; def VLD1_X3 : WInst<"vld1_x3", "3(c*!)", - "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">; + "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPsmQm">; def VLD1_X4 : WInst<"vld1_x4", "4(c*!)", - "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPs">; + "cfilsUcUiUlUsQcQfQiQlQsQUcQUiQUlQUsPcPsQPcQPsmQm">; def VLD1_LANE : WInst<"vld1_lane", ".(c*!).I", - "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs", + "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPsmQm", [ImmCheck<2, ImmCheckLaneIndex, 1>]>; def VLD1_DUP : WInst<"vld1_dup", ".(c*!)", - "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">; + "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPsmQm">; def VST1 : WInst<"vst1", "v*(.!)", "QUcQUsQUiQUlQcQsQiQlQfQPcQPsUcUsUiUlcsilfPcPs">; def VST1_X2 : WInst<"vst1_x2", "v*(2!)", @@ -495,20 +495,20 @@ def VST1_LANE_F16 : WInst<"vst1_lane", "v*(.!)I", "hQh", //////////////////////////////////////////////////////////////////////////////// // E.3.15 Loads and stores of an N-element structure -def VLD2 : WInst<"vld2", "2(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">; -def VLD3 : WInst<"vld3", "3(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">; -def VLD4 : WInst<"vld4", "4(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">; +def VLD2 : WInst<"vld2", "2(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPsmQm">; +def VLD3 : WInst<"vld3", "3(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPsmQm">; +def VLD4 : WInst<"vld4", "4(c*!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPsmQm">; def VLD2_DUP : WInst<"vld2_dup", "2(c*!)", - "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">; + "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUsmQm">; def VLD3_DUP : WInst<"vld3_dup", "3(c*!)", - "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">; + "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUsmQm">; def VLD4_DUP : WInst<"vld4_dup", "4(c*!)", - "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUs">; -def VLD2_LANE : WInst<"vld2_lane", "2(c*!)2I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs", + "UcUsUiUlcsilfPcPsQcQfQiQlQsQPcQPsQUcQUiQUlQUsmQm">; +def VLD2_LANE : WInst<"vld2_lane", "2(c*!)2I", "QUsQUiQsQiQfQPsUcUsUicsifPcPsmQm", [ImmCheck<4, ImmCheckLaneIndex, 1>]>; -def VLD3_LANE : WInst<"vld3_lane", "3(c*!)3I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs", +def VLD3_LANE : WInst<"vld3_lane", "3(c*!)3I", "QUsQUiQsQiQfQPsUcUsUicsifPcPsmQm", [ImmCheck<5, ImmCheckLaneIndex, 1>]>; -def VLD4_LANE : WInst<"vld4_lane", "4(c*!)4I", "QUsQUiQsQiQfQPsUcUsUicsifPcPs", +def VLD4_LANE : WInst<"vld4_lane", "4(c*!)4I", "QUsQUiQsQiQfQPsUcUsUicsifPcPsmQm", [ImmCheck<6, ImmCheckLaneIndex, 1>]>; def VST2 : WInst<"vst2", "v*(2!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">; def VST3 : WInst<"vst3", "v*(3!)", "QUcQUsQUiQcQsQiQfQPcQPsUcUsUiUlcsilfPcPs">; diff --git a/clang/test/CodeGen/AArch64/neon-fp8-intrinsics/acle_neon_loads.c b/clang/test/CodeGen/AArch64/neon-fp8-intrinsics/acle_neon_loads.c new file mode 100644 index 0000000000000..7947494a9dbb8 --- /dev/null +++ b/clang/test/CodeGen/AArch64/neon-fp8-intrinsics/acle_neon_loads.c @@ -0,0 +1,825 @@ +// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5 +// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +neon -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,sroa | FileCheck %s +// RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +neon -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,sroa | FileCheck %s -check-prefix CHECK-CXX + +// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +neon -Werror -Wall -S -O3 -o /dev/null %s + +// REQUIRES: aarch64-registered-target + +#include <arm_neon.h> + +// CHECK-LABEL: define dso_local <8 x i8> @test_vld1_mf8( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = load <8 x i8>, ptr [[A]], align 1 +// CHECK-NEXT: ret <8 x i8> [[TMP0]] +// +// CHECK-CXX-LABEL: define dso_local <8 x i8> @_Z13test_vld1_mf8PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0:[0-9]+]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[TMP0:%.*]] = load <8 x i8>, ptr [[A]], align 1 +// CHECK-CXX-NEXT: ret <8 x i8> [[TMP0]] +// +mfloat8x8_t test_vld1_mf8(mfloat8_t const *a) { + return vld1_mf8(a); +} + +// CHECK-LABEL: define dso_local %struct.mfloat8x8x2_t @test_vld1_mf8_x2( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x2.v8i8.p0(ptr [[A]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-NEXT: ret [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_1_INSERT]] +// +// CHECK-CXX-LABEL: define dso_local %struct.mfloat8x8x2_t @_Z16test_vld1_mf8_x2PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x2.v8i8.p0(ptr [[A]]) +// CHECK-CXX-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-CXX-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-CXX-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-CXX-NEXT: ret [[STRUCT_MFLOAT8X8X2_T]] [[DOTFCA_0_1_INSERT]] +// +mfloat8x8x2_t test_vld1_mf8_x2(mfloat8_t const *a) { + return vld1_mf8_x2(a); +} + +// CHECK-LABEL: define dso_local %struct.mfloat8x8x3_t @test_vld1_mf8_x3( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x3.v8i8.p0(ptr [[A]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 2 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_1_INSERT]], <8 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-NEXT: ret [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_2_INSERT]] +// +// CHECK-CXX-LABEL: define dso_local %struct.mfloat8x8x3_t @_Z16test_vld1_mf8_x3PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x3.v8i8.p0(ptr [[A]]) +// CHECK-CXX-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 2 +// CHECK-CXX-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-CXX-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-CXX-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_1_INSERT]], <8 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-CXX-NEXT: ret [[STRUCT_MFLOAT8X8X3_T]] [[DOTFCA_0_2_INSERT]] +// +mfloat8x8x3_t test_vld1_mf8_x3(mfloat8_t const *a) { + return vld1_mf8_x3(a); +} + +// CHECK-LABEL: define dso_local %struct.mfloat8x8x4_t @test_vld1_mf8_x4( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x4.v8i8.p0(ptr [[A]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 2 +// CHECK-NEXT: [[VLD1XN_FCA_3_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 3 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_1_INSERT]], <8 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-NEXT: [[DOTFCA_0_3_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_2_INSERT]], <8 x i8> [[VLD1XN_FCA_3_EXTRACT]], 0, 3 +// CHECK-NEXT: ret [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_3_INSERT]] +// +// CHECK-CXX-LABEL: define dso_local %struct.mfloat8x8x4_t @_Z16test_vld1_mf8_x4PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[VLD1XN:%.*]] = call { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } @llvm.aarch64.neon.ld1x4.v8i8.p0(ptr [[A]]) +// CHECK-CXX-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 0 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 1 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 2 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_3_EXTRACT:%.*]] = extractvalue { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> } [[VLD1XN]], 3 +// CHECK-CXX-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T:%.*]] poison, <8 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-CXX-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_0_INSERT]], <8 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-CXX-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_1_INSERT]], <8 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-CXX-NEXT: [[DOTFCA_0_3_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_2_INSERT]], <8 x i8> [[VLD1XN_FCA_3_EXTRACT]], 0, 3 +// CHECK-CXX-NEXT: ret [[STRUCT_MFLOAT8X8X4_T]] [[DOTFCA_0_3_INSERT]] +// +mfloat8x8x4_t test_vld1_mf8_x4(mfloat8_t const *a) { + return vld1_mf8_x4(a); +} + +// CHECK-LABEL: define dso_local <8 x i8> @test_vld1_dup_mf8( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[A]], align 1 +// CHECK-NEXT: [[TMP1:%.*]] = insertelement <8 x i8> poison, i8 [[TMP0]], i32 0 +// CHECK-NEXT: [[LANE:%.*]] = shufflevector <8 x i8> [[TMP1]], <8 x i8> [[TMP1]], <8 x i32> zeroinitializer +// CHECK-NEXT: ret <8 x i8> [[LANE]] +// +// CHECK-CXX-LABEL: define dso_local <8 x i8> @_Z17test_vld1_dup_mf8Pu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[TMP0:%.*]] = load i8, ptr [[A]], align 1 +// CHECK-CXX-NEXT: [[TMP1:%.*]] = insertelement <8 x i8> poison, i8 [[TMP0]], i32 0 +// CHECK-CXX-NEXT: [[LANE:%.*]] = shufflevector <8 x i8> [[TMP1]], <8 x i8> [[TMP1]], <8 x i32> zeroinitializer +// CHECK-CXX-NEXT: ret <8 x i8> [[LANE]] +// +mfloat8x8_t test_vld1_dup_mf8(mfloat8_t *a) { + return vld1_dup_mf8(a); +} + +// CHECK-LABEL: define dso_local <8 x i8> @test_vld1_lane_mf8( +// CHECK-SAME: ptr noundef [[A:%.*]], <8 x i8> [[B:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[A]], align 1 +// CHECK-NEXT: [[VLD1_LANE:%.*]] = insertelement <8 x i8> [[B]], i8 [[TMP0]], i32 7 +// CHECK-NEXT: ret <8 x i8> [[VLD1_LANE]] +// +// CHECK-CXX-LABEL: define dso_local <8 x i8> @_Z18test_vld1_lane_mf8Pu6__mfp813__Mfloat8x8_t( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]], <8 x i8> [[B:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[TMP0:%.*]] = load i8, ptr [[A]], align 1 +// CHECK-CXX-NEXT: [[VLD1_LANE:%.*]] = insertelement <8 x i8> [[B]], i8 [[TMP0]], i32 7 +// CHECK-CXX-NEXT: ret <8 x i8> [[VLD1_LANE]] +// +mfloat8x8_t test_vld1_lane_mf8(mfloat8_t *a, mfloat8x8_t b) { + return vld1_lane_mf8(a, b, 7); +} + +// CHECK-LABEL: define dso_local <16 x i8> @test_vld1q_mf8( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = load <16 x i8>, ptr [[A]], align 1 +// CHECK-NEXT: ret <16 x i8> [[TMP0]] +// +// CHECK-CXX-LABEL: define dso_local <16 x i8> @_Z14test_vld1q_mf8PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[TMP0:%.*]] = load <16 x i8>, ptr [[A]], align 1 +// CHECK-CXX-NEXT: ret <16 x i8> [[TMP0]] +// +mfloat8x16_t test_vld1q_mf8(mfloat8_t const *a) { + return vld1q_mf8(a); +} + +// CHECK-LABEL: define dso_local %struct.mfloat8x16x2_t @test_vld1q_mf8_x2( +// CHECK-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <16 x i8>, <16 x i8> } @llvm.aarch64.neon.ld1x2.v16i8.p0(ptr [[A]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T:%.*]] poison, <16 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-NEXT: ret [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_1_INSERT]] +// +// CHECK-CXX-LABEL: define dso_local %struct.mfloat8x16x2_t @_Z17test_vld1q_mf8_x2PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[A:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[VLD1XN:%.*]] = call { <16 x i8>, <16 x i8> } @llvm.aarch64.neon.ld1x2.v16i8.p0(ptr [[A]]) +// CHECK-CXX-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[VLD1XN]], 0 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[VLD1XN]], 1 +// CHECK-CXX-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T:%.*]] poison, <16 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-CXX-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-CXX-NEXT: ret [[STRUCT_MFLOAT8X16X2_T]] [[DOTFCA_0_1_INSERT]] +// +mfloat8x16x2_t test_vld1q_mf8_x2(mfloat8_t const *a) { + return vld1q_mf8_x2(a); +} + +// CHECK-LABEL: define dso_local %struct.mfloat8x16x3_t @test_vld1q_mf8_x3( +// CHECK-SAME: ptr noundef [[PTR:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <16 x i8>, <16 x i8>, <16 x i8> } @llvm.aarch64.neon.ld1x3.v16i8.p0(ptr [[PTR]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 2 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T:%.*]] poison, <16 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_1_INSERT]], <16 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-NEXT: ret [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_2_INSERT]] +// +// CHECK-CXX-LABEL: define dso_local %struct.mfloat8x16x3_t @_Z17test_vld1q_mf8_x3PKu6__mfp8( +// CHECK-CXX-SAME: ptr noundef [[PTR:%.*]]) #[[ATTR0]] { +// CHECK-CXX-NEXT: [[ENTRY:.*:]] +// CHECK-CXX-NEXT: [[VLD1XN:%.*]] = call { <16 x i8>, <16 x i8>, <16 x i8> } @llvm.aarch64.neon.ld1x3.v16i8.p0(ptr [[PTR]]) +// CHECK-CXX-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 0 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 1 +// CHECK-CXX-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 2 +// CHECK-CXX-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T:%.*]] poison, <16 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-CXX-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[VLD1XN_FCA_1_EXTRACT]], 0, 1 +// CHECK-CXX-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_1_INSERT]], <16 x i8> [[VLD1XN_FCA_2_EXTRACT]], 0, 2 +// CHECK-CXX-NEXT: ret [[STRUCT_MFLOAT8X16X3_T]] [[DOTFCA_0_2_INSERT]] +// +mfloat8x16x3_t test_vld1q_mf8_x3(mfloat8_t const *ptr) { + return vld1q_mf8_x3(ptr); +} +// CHECK-LABEL: define dso_local %struct.mfloat8x16x4_t @test_vld1q_mf8_x4( +// CHECK-SAME: ptr noundef [[PTR:%.*]]) #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[VLD1XN:%.*]] = call { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } @llvm.aarch64.neon.ld1x4.v16i8.p0(ptr [[PTR]]) +// CHECK-NEXT: [[VLD1XN_FCA_0_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 0 +// CHECK-NEXT: [[VLD1XN_FCA_1_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 1 +// CHECK-NEXT: [[VLD1XN_FCA_2_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 2 +// CHECK-NEXT: [[VLD1XN_FCA_3_EXTRACT:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[VLD1XN]], 3 +// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X4_T:%.*]] poison, <16 x i8> [[VLD1XN_FCA_0_EXTRACT]], 0, 0 +// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_MFLOAT8X16X4_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[VLD1XN_FCA_1_EXTR... [truncated] `````````` </details> https://github.com/llvm/llvm-project/pull/145666 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
