clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp

   1 // RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -S -O1 -Werror -Wall -emit-llvm -o - %s -mvscale-min=1 -mvscale-max=1  | FileCheck %s -D#VBITS=128  --check-prefixes=CHECK,CHECK128
   2 // RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -S -O1 -Werror -Wall -emit-llvm -o - %s -mvscale-min=2 -mvscale-max=2  | FileCheck %s -D#VBITS=256  --check-prefixes=CHECK,CHECKWIDE
   3 // RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -S -O1 -Werror -Wall -emit-llvm -o - %s -mvscale-min=4 -mvscale-max=4  | FileCheck %s -D#VBITS=512  --check-prefixes=CHECK,CHECKWIDE
   4 // RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -S -O1 -Werror -Wall -emit-llvm -o - %s -mvscale-min=8 -mvscale-max=8 | FileCheck %s -D#VBITS=1024 --check-prefixes=CHECK,CHECKWIDE
   5 // RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -S -O1 -Werror -Wall -emit-llvm -o - %s -mvscale-min=16 -mvscale-max=16 | FileCheck %s -D#VBITS=2048 --check-prefixes=CHECK,CHECKWIDE
   6
   7 // REQUIRES: aarch64-registered-target
   8
   9 // Examples taken from section "3.7.3.3 Behavior specific to SVE
  10 // vectors" of the SVE ACLE (Version 00bet6) that can be found at
  11 // https://developer.arm.com/documentation/100987/latest
  12 //
  13 // Example has been expanded to work with mutiple values of
  14 // -mvscale-{min,max}.
  15
  16 #include <arm_sve.h>
  17
  18 // Page 26, first paragraph of 3.7.3.3: sizeof and alignof
  19 #if __ARM_FEATURE_SVE_BITS
  20 #define N __ARM_FEATURE_SVE_BITS
  21 typedef svfloat32_t fixed_svfloat __attribute__((arm_sve_vector_bits(N)));
  22 void test01() {
  23   static_assert(alignof(fixed_svfloat) == 16,
  24                 "Invalid align of Vector Length Specific Type.");
  25   static_assert(sizeof(fixed_svfloat) == N / 8,
  26                 "Invalid size of Vector Length Specific Type.");
  27 }
  28 #endif
  29
  30 // Page 26, items 1 and 2 of 3.7.3.3: how VLST and GNUT are related.
  31 #if __ARM_FEATURE_SVE_BITS && __ARM_FEATURE_SVE_VECTOR_OPERATORS
  32 #define N __ARM_FEATURE_SVE_BITS
  33 typedef svfloat64_t fixed_svfloat64 __attribute__((arm_sve_vector_bits(N)));
  34 typedef float64_t gnufloat64 __attribute__((vector_size(N / 8)));
  35 void test02() {
  36   static_assert(alignof(fixed_svfloat64) == alignof(gnufloat64),
  37                 "Align of Vector Length Specific Type and GNU Vector Types "
  38                 "should be the same.");
  39   static_assert(sizeof(fixed_svfloat64) == sizeof(gnufloat64),
  40                 "Size of Vector Length Specific Type and GNU Vector Types "
  41                 "should be the same.");
  42 }
  43 #endif
  44
  45 // Page 27, item 1.
  46 #if __ARM_FEATURE_SVE_BITS && __ARM_FEATURE_SVE_VECTOR_OPERATORS
  47 #define N __ARM_FEATURE_SVE_BITS
  48 // CHECK-LABEL: define{{.*}} <vscale x 4 x i32> @_Z1f9__SVE_VLSIu11__SVInt32_tLj
  49 // CHECK-SAME:    [[#VBITS]]
  50 // CHECK-SAME:    EES_(<vscale x 4 x i32> noundef %x.coerce, <vscale x 4 x i32> noundef %y.coerce)
  51 // CHECK-NEXT: entry:
  52 // CHECK-NEXT:   [[X:%.*]] = tail call <[[#div(VBITS, 32)]] x i32> @llvm.vector.extract.v[[#div(VBITS, 32)]]i32.nxv4i32(<vscale x 4 x i32> [[X_COERCE:%.*]], i64 0)
  53 // CHECK-NEXT:   [[Y:%.*]] = tail call <[[#div(VBITS, 32)]] x i32> @llvm.vector.extract.v[[#div(VBITS, 32)]]i32.nxv4i32(<vscale x 4 x i32> [[X_COERCE1:%.*]], i64 0)
  54 // CHECK-NEXT:   [[ADD:%.*]] = add <[[#div(VBITS, 32)]] x i32> [[Y]], [[X]]
  55 // CHECK-NEXT:   [[CASTSCALABLESVE:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v[[#div(VBITS, 32)]]i32(<vscale x 4 x i32> undef, <[[#div(VBITS, 32)]] x i32> [[ADD]], i64 0)
  56 // CHECK-NEXT:   ret <vscale x 4 x i32> [[CASTSCALABLESVE]]
  57 typedef svint32_t vec __attribute__((arm_sve_vector_bits(N)));
  58 auto f(vec x, vec y) { return x + y; } // Returns a vec.
  59 #endif
  60
  61 // Page 27, item 3, adapted for a generic value of __ARM_FEATURE_SVE_BITS
  62 #if __ARM_FEATURE_SVE_BITS && __ARM_FEATURE_SVE_VECTOR_OPERATORS
  63 #define N __ARM_FEATURE_SVE_BITS
  64 typedef int16_t vec1 __attribute__((vector_size(N / 8)));
  65 void f(vec1);
  66 typedef svint16_t vec2 __attribute__((arm_sve_vector_bits(N)));
  67 // CHECK-LABEL: define{{.*}} void @_Z1g9__SVE_VLSIu11__SVInt16_tLj
  68 // CHECK-SAME:    [[#VBITS]]
  69 // CHECK-SAME:    EE(<vscale x 8 x i16> noundef %x.coerce)
  70 // CHECK-NEXT: entry:
  71 // CHECK128-NEXT:   [[X:%.*]] = tail call <8 x i16> @llvm.vector.extract.v8i16.nxv8i16(<vscale x 8 x i16> [[X_COERCE:%.*]], i64 0)
  72 // CHECK128-NEXT:   call void @_Z1fDv8_s(<8 x i16> noundef [[X]]) [[ATTR5:#.*]]
  73 // CHECK128-NEXT:   ret void
  74 // CHECKWIDE-NEXT:   [[INDIRECT_ARG_TEMP:%.*]] = alloca <[[#div(VBITS, 16)]] x i16>, align 16
  75 // CHECKWIDE-NEXT:   [[X:%.*]] = tail call <[[#div(VBITS, 16)]] x i16> @llvm.vector.extract.v[[#div(VBITS, 16)]]i16.nxv8i16(<vscale x 8 x i16> [[X_COERCE:%.*]], i64 0)
  76 // CHECKWIDE-NEXT:   store <[[#div(VBITS, 16)]] x i16> [[X]], ptr [[INDIRECT_ARG_TEMP]], align 16, [[TBAA6:!tbaa !.*]]
  77 // CHECKWIDE-NEXT:   call void @_Z1fDv[[#div(VBITS, 16)]]_s(ptr noundef nonnull [[INDIRECT_ARG_TEMP]]) [[ATTR5:#.*]]
  78 // CHECKWIDE-NEXT:   ret void
  79 void g(vec2 x) { f(x); } // OK
  80 #endif