From: Sanjay Patel Date: Wed, 15 Jun 2016 17:17:27 +0000 (+0000) Subject: [x86, SSE] remove the GCCBuiltins from the integer min/max intrinsics X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=3539217c1c6c5f94fc8fc8b38b1ea1b44f52f034;p=llvm [x86, SSE] remove the GCCBuiltins from the integer min/max intrinsics This allows us to emit native IR in Clang (next commit). Also, update the intrinsic tests to show that codegen already knows how to handle the IR that Clang will soon produce. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@272806 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td index 983840a4bfd..75b0d64643a 100644 --- a/include/llvm/IR/IntrinsicsX86.td +++ b/include/llvm/IR/IntrinsicsX86.td @@ -406,16 +406,16 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_sse2_pavg_w : GCCBuiltin<"__builtin_ia32_pavgw128">, Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem, Commutative]>; - def int_x86_sse2_pmaxu_b : GCCBuiltin<"__builtin_ia32_pmaxub128">, + def int_x86_sse2_pmaxu_b : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem, Commutative]>; - def int_x86_sse2_pmaxs_w : GCCBuiltin<"__builtin_ia32_pmaxsw128">, + def int_x86_sse2_pmaxs_w : Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem, Commutative]>; - def int_x86_sse2_pminu_b : GCCBuiltin<"__builtin_ia32_pminub128">, + def int_x86_sse2_pminu_b : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem, Commutative]>; - def int_x86_sse2_pmins_w : GCCBuiltin<"__builtin_ia32_pminsw128">, + def int_x86_sse2_pmins_w : Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem, Commutative]>; def int_x86_sse2_psad_bw : GCCBuiltin<"__builtin_ia32_psadbw128">, @@ -737,28 +737,28 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". // Vector compare, min, max let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". - def int_x86_sse41_pmaxsb : GCCBuiltin<"__builtin_ia32_pmaxsb128">, + def int_x86_sse41_pmaxsb : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pmaxsd : GCCBuiltin<"__builtin_ia32_pmaxsd128">, + def int_x86_sse41_pmaxsd : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pmaxud : GCCBuiltin<"__builtin_ia32_pmaxud128">, + def int_x86_sse41_pmaxud : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pmaxuw : GCCBuiltin<"__builtin_ia32_pmaxuw128">, + def int_x86_sse41_pmaxuw : Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pminsb : GCCBuiltin<"__builtin_ia32_pminsb128">, + def int_x86_sse41_pminsb : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pminsd : GCCBuiltin<"__builtin_ia32_pminsd128">, + def int_x86_sse41_pminsd : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pminud : GCCBuiltin<"__builtin_ia32_pminud128">, + def int_x86_sse41_pminud : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem, Commutative]>; - def int_x86_sse41_pminuw : GCCBuiltin<"__builtin_ia32_pminuw128">, + def int_x86_sse41_pminuw : Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty], [IntrNoMem, Commutative]>; } diff --git a/test/CodeGen/X86/sse2-intrinsics-fast-isel.ll b/test/CodeGen/X86/sse2-intrinsics-fast-isel.ll index c8927f8200d..0a961e83d5a 100644 --- a/test/CodeGen/X86/sse2-intrinsics-fast-isel.ll +++ b/test/CodeGen/X86/sse2-intrinsics-fast-isel.ll @@ -1598,8 +1598,9 @@ define <2 x i64> @test_mm_max_epi16(<2 x i64> %a0, <2 x i64> %a1) nounwind { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <8 x i16> %arg1 = bitcast <2 x i64> %a1 to <8 x i16> - %res = call <8 x i16> @llvm.x86.sse2.pmaxs.w(<8 x i16> %arg0, <8 x i16> %arg1) - %bc = bitcast <8 x i16> %res to <2 x i64> + %cmp = icmp sgt <8 x i16> %arg0, %arg1 + %sel = select <8 x i1> %cmp, <8 x i16> %arg0, <8 x i16> %arg1 + %bc = bitcast <8 x i16> %sel to <2 x i64> ret <2 x i64> %bc } declare <8 x i16> @llvm.x86.sse2.pmaxs.w(<8 x i16>, <8 x i16>) nounwind readnone @@ -1616,8 +1617,9 @@ define <2 x i64> @test_mm_max_epu8(<2 x i64> %a0, <2 x i64> %a1) nounwind { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <16 x i8> %arg1 = bitcast <2 x i64> %a1 to <16 x i8> - %res = call <16 x i8> @llvm.x86.sse2.pmaxu.b(<16 x i8> %arg0, <16 x i8> %arg1) - %bc = bitcast <16 x i8> %res to <2 x i64> + %cmp = icmp ugt <16 x i8> %arg0, %arg1 + %sel = select <16 x i1> %cmp, <16 x i8> %arg0, <16 x i8> %arg1 + %bc = bitcast <16 x i8> %sel to <2 x i64> ret <2 x i64> %bc } declare <16 x i8> @llvm.x86.sse2.pmaxu.b(<16 x i8>, <16 x i8>) nounwind readnone @@ -1679,8 +1681,9 @@ define <2 x i64> @test_mm_min_epi16(<2 x i64> %a0, <2 x i64> %a1) nounwind { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <8 x i16> %arg1 = bitcast <2 x i64> %a1 to <8 x i16> - %res = call <8 x i16> @llvm.x86.sse2.pmins.w(<8 x i16> %arg0, <8 x i16> %arg1) - %bc = bitcast <8 x i16> %res to <2 x i64> + %cmp = icmp slt <8 x i16> %arg0, %arg1 + %sel = select <8 x i1> %cmp, <8 x i16> %arg0, <8 x i16> %arg1 + %bc = bitcast <8 x i16> %sel to <2 x i64> ret <2 x i64> %bc } declare <8 x i16> @llvm.x86.sse2.pmins.w(<8 x i16>, <8 x i16>) nounwind readnone @@ -1697,8 +1700,9 @@ define <2 x i64> @test_mm_min_epu8(<2 x i64> %a0, <2 x i64> %a1) nounwind { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <16 x i8> %arg1 = bitcast <2 x i64> %a1 to <16 x i8> - %res = call <16 x i8> @llvm.x86.sse2.pminu.b(<16 x i8> %arg0, <16 x i8> %arg1) - %bc = bitcast <16 x i8> %res to <2 x i64> + %cmp = icmp ult <16 x i8> %arg0, %arg1 + %sel = select <16 x i1> %cmp, <16 x i8> %arg0, <16 x i8> %arg1 + %bc = bitcast <16 x i8> %sel to <2 x i64> ret <2 x i64> %bc } declare <16 x i8> @llvm.x86.sse2.pminu.b(<16 x i8>, <16 x i8>) nounwind readnone diff --git a/test/CodeGen/X86/sse41-intrinsics-fast-isel.ll b/test/CodeGen/X86/sse41-intrinsics-fast-isel.ll index 7bfce0941a2..f1e01db5960 100644 --- a/test/CodeGen/X86/sse41-intrinsics-fast-isel.ll +++ b/test/CodeGen/X86/sse41-intrinsics-fast-isel.ll @@ -602,11 +602,11 @@ define <2 x i64> @test_mm_max_epi8(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <16 x i8> %arg1 = bitcast <2 x i64> %a1 to <16 x i8> - %res = call <16 x i8> @llvm.x86.sse41.pmaxsb(<16 x i8> %arg0, <16 x i8> %arg1) - %bc = bitcast <16 x i8> %res to <2 x i64> + %cmp = icmp sgt <16 x i8> %arg0, %arg1 + %sel = select <16 x i1> %cmp, <16 x i8> %arg0, <16 x i8> %arg1 + %bc = bitcast <16 x i8> %sel to <2 x i64> ret <2 x i64> %bc } -declare <16 x i8> @llvm.x86.sse41.pmaxsb(<16 x i8>, <16 x i8>) nounwind readnone define <2 x i64> @test_mm_max_epi32(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_max_epi32: @@ -620,11 +620,11 @@ define <2 x i64> @test_mm_max_epi32(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <4 x i32> %arg1 = bitcast <2 x i64> %a1 to <4 x i32> - %res = call <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32> %arg0, <4 x i32> %arg1) - %bc = bitcast <4 x i32> %res to <2 x i64> + %cmp = icmp sgt <4 x i32> %arg0, %arg1 + %sel = select <4 x i1> %cmp, <4 x i32> %arg0, <4 x i32> %arg1 + %bc = bitcast <4 x i32> %sel to <2 x i64> ret <2 x i64> %bc } -declare <4 x i32> @llvm.x86.sse41.pmaxsd(<4 x i32>, <4 x i32>) nounwind readnone define <2 x i64> @test_mm_max_epu16(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_max_epu16: @@ -638,11 +638,11 @@ define <2 x i64> @test_mm_max_epu16(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <8 x i16> %arg1 = bitcast <2 x i64> %a1 to <8 x i16> - %res = call <8 x i16> @llvm.x86.sse41.pmaxuw(<8 x i16> %arg0, <8 x i16> %arg1) - %bc = bitcast <8 x i16> %res to <2 x i64> + %cmp = icmp ugt <8 x i16> %arg0, %arg1 + %sel = select <8 x i1> %cmp, <8 x i16> %arg0, <8 x i16> %arg1 + %bc = bitcast <8 x i16> %sel to <2 x i64> ret <2 x i64> %bc } -declare <8 x i16> @llvm.x86.sse41.pmaxuw(<8 x i16>, <8 x i16>) nounwind readnone define <2 x i64> @test_mm_max_epu32(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_max_epu32: @@ -656,11 +656,11 @@ define <2 x i64> @test_mm_max_epu32(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <4 x i32> %arg1 = bitcast <2 x i64> %a1 to <4 x i32> - %res = call <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32> %arg0, <4 x i32> %arg1) - %bc = bitcast <4 x i32> %res to <2 x i64> + %cmp = icmp ugt <4 x i32> %arg0, %arg1 + %sel = select <4 x i1> %cmp, <4 x i32> %arg0, <4 x i32> %arg1 + %bc = bitcast <4 x i32> %sel to <2 x i64> ret <2 x i64> %bc } -declare <4 x i32> @llvm.x86.sse41.pmaxud(<4 x i32>, <4 x i32>) nounwind readnone define <2 x i64> @test_mm_min_epi8(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_min_epi8: @@ -674,11 +674,11 @@ define <2 x i64> @test_mm_min_epi8(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <16 x i8> %arg1 = bitcast <2 x i64> %a1 to <16 x i8> - %res = call <16 x i8> @llvm.x86.sse41.pminsb(<16 x i8> %arg0, <16 x i8> %arg1) - %bc = bitcast <16 x i8> %res to <2 x i64> + %cmp = icmp slt <16 x i8> %arg0, %arg1 + %sel = select <16 x i1> %cmp, <16 x i8> %arg0, <16 x i8> %arg1 + %bc = bitcast <16 x i8> %sel to <2 x i64> ret <2 x i64> %bc } -declare <16 x i8> @llvm.x86.sse41.pminsb(<16 x i8>, <16 x i8>) nounwind readnone define <2 x i64> @test_mm_min_epi32(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_min_epi32: @@ -692,11 +692,11 @@ define <2 x i64> @test_mm_min_epi32(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <4 x i32> %arg1 = bitcast <2 x i64> %a1 to <4 x i32> - %res = call <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32> %arg0, <4 x i32> %arg1) - %bc = bitcast <4 x i32> %res to <2 x i64> + %cmp = icmp slt <4 x i32> %arg0, %arg1 + %sel = select <4 x i1> %cmp, <4 x i32> %arg0, <4 x i32> %arg1 + %bc = bitcast <4 x i32> %sel to <2 x i64> ret <2 x i64> %bc } -declare <4 x i32> @llvm.x86.sse41.pminsd(<4 x i32>, <4 x i32>) nounwind readnone define <2 x i64> @test_mm_min_epu16(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_min_epu16: @@ -710,11 +710,11 @@ define <2 x i64> @test_mm_min_epu16(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <8 x i16> %arg1 = bitcast <2 x i64> %a1 to <8 x i16> - %res = call <8 x i16> @llvm.x86.sse41.pminuw(<8 x i16> %arg0, <8 x i16> %arg1) - %bc = bitcast <8 x i16> %res to <2 x i64> + %cmp = icmp ult <8 x i16> %arg0, %arg1 + %sel = select <8 x i1> %cmp, <8 x i16> %arg0, <8 x i16> %arg1 + %bc = bitcast <8 x i16> %sel to <2 x i64> ret <2 x i64> %bc } -declare <8 x i16> @llvm.x86.sse41.pminuw(<8 x i16>, <8 x i16>) nounwind readnone define <2 x i64> @test_mm_min_epu32(<2 x i64> %a0, <2 x i64> %a1) { ; X32-LABEL: test_mm_min_epu32: @@ -728,11 +728,11 @@ define <2 x i64> @test_mm_min_epu32(<2 x i64> %a0, <2 x i64> %a1) { ; X64-NEXT: retq %arg0 = bitcast <2 x i64> %a0 to <4 x i32> %arg1 = bitcast <2 x i64> %a1 to <4 x i32> - %res = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %arg0, <4 x i32> %arg1) - %bc = bitcast <4 x i32> %res to <2 x i64> + %cmp = icmp ult <4 x i32> %arg0, %arg1 + %sel = select <4 x i1> %cmp, <4 x i32> %arg0, <4 x i32> %arg1 + %bc = bitcast <4 x i32> %sel to <2 x i64> ret <2 x i64> %bc } -declare <4 x i32> @llvm.x86.sse41.pminud(<4 x i32>, <4 x i32>) nounwind readnone define <2 x i64> @test_mm_minpos_epu16(<2 x i64> %a0) { ; X32-LABEL: test_mm_minpos_epu16: