clang/test/CodeGen/fp-matrix-pragma.c

   1 // RUN: %clang -emit-llvm -S -fenable-matrix -mllvm -disable-llvm-optzns %s -o - | FileCheck %s
   2
   3 typedef float fx2x2_t __attribute__((matrix_type(2, 2)));
   4 typedef int ix2x2_t __attribute__((matrix_type(2, 2)));
   5
   6 fx2x2_t fp_matrix_contract(fx2x2_t a, fx2x2_t b, float c, float d) {
   7 // CHECK: call contract <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32
   8 // CHECK: fdiv contract <4 x float>
   9 // CHECK: fmul contract <4 x float>
  10 #pragma clang fp contract(fast)
  11   return (a * b / c) * d;
  12 }
  13
  14 fx2x2_t fp_matrix_reassoc(fx2x2_t a, fx2x2_t b, fx2x2_t c) {
  15 // CHECK: fadd reassoc <4 x float>
  16 // CHECK: fsub reassoc <4 x float>
  17 #pragma clang fp reassociate(on)
  18   return a + b - c;
  19 }
  20
  21 fx2x2_t fp_matrix_ops(fx2x2_t a, fx2x2_t b, fx2x2_t c) {
  22 // CHECK: call reassoc contract <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32
  23 // CHECK: fadd reassoc contract <4 x float>
  24 #pragma clang fp contract(fast) reassociate(on)
  25   return a * b + c;
  26 }
  27
  28 fx2x2_t fp_matrix_compound_ops(fx2x2_t a, fx2x2_t b, fx2x2_t c, fx2x2_t d,
  29     float e, float f) {
  30 // CHECK: call reassoc contract <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32
  31 // CHECK: fadd reassoc contract <4 x float>
  32 // CHECK: fsub reassoc contract <4 x float>
  33 // CHECK: fmul reassoc contract <4 x float>
  34 // CHECK: fdiv reassoc contract <4 x float>
  35 #pragma clang fp contract(fast) reassociate(on)
  36   a *= b;
  37   a += c;
  38   a -= d;
  39   a *= e;
  40   a /= f;
  41
  42   return a;
  43 }
  44
  45 ix2x2_t int_matrix_ops(ix2x2_t a, ix2x2_t b, ix2x2_t c) {
  46 // CHECK: call <4 x i32> @llvm.matrix.multiply.v4i32.v4i32.v4i32
  47 // CHECK: add <4 x i32>
  48 #pragma clang fp contract(fast) reassociate(on)
  49   return a * b + c;
  50 }