From 6c6344e6a95130b61ba8491b5db262e8a171f30b Mon Sep 17 00:00:00 2001 From: Chad Rosier Date: Mon, 9 Dec 2013 22:47:31 +0000 Subject: [AArch64] Remove q and non-q intrinsic definitions in the NEON scalar reduce pairwise implementation, using an overloaded definition instead. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196831 91177308-0d34-0410-b5e6-96231b3b80d8 --- .../CodeGen/AArch64/neon-scalar-reduce-pairwise.ll | 87 +++++++++++----------- 1 file changed, 43 insertions(+), 44 deletions(-) (limited to 'test/CodeGen') diff --git a/test/CodeGen/AArch64/neon-scalar-reduce-pairwise.ll b/test/CodeGen/AArch64/neon-scalar-reduce-pairwise.ll index 80e8dc339d..401ceec255 100644 --- a/test/CodeGen/AArch64/neon-scalar-reduce-pairwise.ll +++ b/test/CodeGen/AArch64/neon-scalar-reduce-pairwise.ll @@ -4,101 +4,100 @@ declare <1 x i64> @llvm.aarch64.neon.vpadd(<2 x i64>) define <1 x i64> @test_addp_v1i64(<2 x i64> %a) { ; CHECK: test_addp_v1i64: - %val = call <1 x i64> @llvm.aarch64.neon.vpadd(<2 x i64> %a) -; CHECK: addp d0, v0.2d - ret <1 x i64> %val +; CHECK: addp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x i64> @llvm.aarch64.neon.vpadd(<2 x i64> %a) + ret <1 x i64> %val } -declare <1 x float> @llvm.aarch64.neon.vpfadd(<2 x float>) +declare <1 x float> @llvm.aarch64.neon.vpfadd.v1f32.v2f32(<2 x float>) define <1 x float> @test_faddp_v1f32(<2 x float> %a) { ; CHECK: test_faddp_v1f32: - %val = call <1 x float> @llvm.aarch64.neon.vpfadd(<2 x float> %a) -; CHECK: faddp s0, v0.2s - ret <1 x float> %val +; CHECK: faddp {{s[0-9]+}}, {{v[0-9]+}}.2s + %val = call <1 x float> @llvm.aarch64.neon.vpfadd.v1f32.v2f32(<2 x float> %a) + ret <1 x float> %val } -declare <1 x double> @llvm.aarch64.neon.vpfaddq(<2 x double>) +declare <1 x double> @llvm.aarch64.neon.vpfadd.v1f64.v2f64(<2 x double>) define <1 x double> @test_faddp_v1f64(<2 x double> %a) { ; CHECK: test_faddp_v1f64: - %val = call <1 x double> @llvm.aarch64.neon.vpfaddq(<2 x double> %a) -; CHECK: faddp d0, v0.2d - ret <1 x double> %val +; CHECK: faddp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x double> @llvm.aarch64.neon.vpfadd.v1f64.v2f64(<2 x double> %a) + ret <1 x double> %val } -declare <1 x float> @llvm.aarch64.neon.vpmax(<2 x float>) +declare <1 x float> @llvm.aarch64.neon.vpmax.v1f32.v2f32(<2 x float>) define <1 x float> @test_fmaxp_v1f32(<2 x float> %a) { ; CHECK: test_fmaxp_v1f32: - %val = call <1 x float> @llvm.aarch64.neon.vpmax(<2 x float> %a) -; CHECK: fmaxp s0, v0.2s - ret <1 x float> %val +; CHECK: fmaxp {{s[0-9]+}}, {{v[0-9]+}}.2s + %val = call <1 x float> @llvm.aarch64.neon.vpmax.v1f32.v2f32(<2 x float> %a) + ret <1 x float> %val } -declare <1 x double> @llvm.aarch64.neon.vpmaxq(<2 x double>) +declare <1 x double> @llvm.aarch64.neon.vpmax.v1f64.v2f64(<2 x double>) define <1 x double> @test_fmaxp_v1f64(<2 x double> %a) { ; CHECK: test_fmaxp_v1f64: - %val = call <1 x double> @llvm.aarch64.neon.vpmaxq(<2 x double> %a) -; CHECK: fmaxp d0, v0.2d - ret <1 x double> %val +; CHECK: fmaxp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x double> @llvm.aarch64.neon.vpmax.v1f64.v2f64(<2 x double> %a) + ret <1 x double> %val } - -declare <1 x float> @llvm.aarch64.neon.vpmin(<2 x float>) +declare <1 x float> @llvm.aarch64.neon.vpmin.v1f32.v2f32(<2 x float>) define <1 x float> @test_fminp_v1f32(<2 x float> %a) { ; CHECK: test_fminp_v1f32: - %val = call <1 x float> @llvm.aarch64.neon.vpmin(<2 x float> %a) -; CHECK: fminp s0, v0.2s - ret <1 x float> %val +; CHECK: fminp {{s[0-9]+}}, {{v[0-9]+}}.2s + %val = call <1 x float> @llvm.aarch64.neon.vpmin.v1f32.v2f32(<2 x float> %a) + ret <1 x float> %val } -declare <1 x double> @llvm.aarch64.neon.vpminq(<2 x double>) +declare <1 x double> @llvm.aarch64.neon.vpmin.v1f64.v2f64(<2 x double>) define <1 x double> @test_fminp_v1f64(<2 x double> %a) { ; CHECK: test_fminp_v1f64: - %val = call <1 x double> @llvm.aarch64.neon.vpminq(<2 x double> %a) -; CHECK: fminp d0, v0.2d - ret <1 x double> %val +; CHECK: fminp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x double> @llvm.aarch64.neon.vpmin.v1f64.v2f64(<2 x double> %a) + ret <1 x double> %val } -declare <1 x float> @llvm.aarch64.neon.vpfmaxnm(<2 x float>) +declare <1 x float> @llvm.aarch64.neon.vpfmaxnm.v1f32.v2f32(<2 x float>) define <1 x float> @test_fmaxnmp_v1f32(<2 x float> %a) { ; CHECK: test_fmaxnmp_v1f32: - %val = call <1 x float> @llvm.aarch64.neon.vpfmaxnm(<2 x float> %a) -; CHECK: fmaxnmp s0, v0.2s - ret <1 x float> %val +; CHECK: fmaxnmp {{s[0-9]+}}, {{v[0-9]+}}.2s + %val = call <1 x float> @llvm.aarch64.neon.vpfmaxnm.v1f32.v2f32(<2 x float> %a) + ret <1 x float> %val } -declare <1 x double> @llvm.aarch64.neon.vpfmaxnmq(<2 x double>) +declare <1 x double> @llvm.aarch64.neon.vpfmaxnm.v1f64.v2f64(<2 x double>) define <1 x double> @test_fmaxnmp_v1f64(<2 x double> %a) { ; CHECK: test_fmaxnmp_v1f64: - %val = call <1 x double> @llvm.aarch64.neon.vpfmaxnmq(<2 x double> %a) -; CHECK: fmaxnmp d0, v0.2d - ret <1 x double> %val +; CHECK: fmaxnmp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x double> @llvm.aarch64.neon.vpfmaxnm.v1f64.v2f64(<2 x double> %a) + ret <1 x double> %val } -declare <1 x float> @llvm.aarch64.neon.vpfminnm(<2 x float>) +declare <1 x float> @llvm.aarch64.neon.vpfminnm.v1f32.v2f32(<2 x float>) define <1 x float> @test_fminnmp_v1f32(<2 x float> %a) { ; CHECK: test_fminnmp_v1f32: - %val = call <1 x float> @llvm.aarch64.neon.vpfminnm(<2 x float> %a) -; CHECK: fminnmp s0, v0.2s - ret <1 x float> %val +; CHECK: fminnmp {{s[0-9]+}}, {{v[0-9]+}}.2s + %val = call <1 x float> @llvm.aarch64.neon.vpfminnm.v1f32.v2f32(<2 x float> %a) + ret <1 x float> %val } -declare <1 x double> @llvm.aarch64.neon.vpfminnmq(<2 x double>) +declare <1 x double> @llvm.aarch64.neon.vpfminnm.v1f64.v2f64(<2 x double>) define <1 x double> @test_fminnmp_v1f64(<2 x double> %a) { ; CHECK: test_fminnmp_v1f64: - %val = call <1 x double> @llvm.aarch64.neon.vpfminnmq(<2 x double> %a) -; CHECK: fminnmp d0, v0.2d - ret <1 x double> %val +; CHECK: fminnmp {{d[0-9]+}}, {{v[0-9]+}}.2d + %val = call <1 x double> @llvm.aarch64.neon.vpfminnm.v1f64.v2f64(<2 x double> %a) + ret <1 x double> %val } define float @test_vaddv_f32(<2 x float> %a) { -- cgit v1.2.3