diff --git a/include/llvm/IntrinsicsX86.td b/include/llvm/IntrinsicsX86.td index 77f787c63bb..aa4c54f8b82 100644 --- a/include/llvm/IntrinsicsX86.td +++ b/include/llvm/IntrinsicsX86.td @@ -265,6 +265,30 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". llvm_int_ty], [IntrNoMem]>; } +// Conversion ops +let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". + def int_x86_sse2_cvtdq2pd : GCCBuiltin<"__builtin_ia32_cvtdq2pd">, + Intrinsic<[llvm_v2f64_ty, llvm_v4i32_ty], [IntrNoMem]>; + def int_x86_sse2_cvtdq2ps : GCCBuiltin<"__builtin_ia32_cvtdq2ps">, + Intrinsic<[llvm_v4f32_ty, llvm_v4i32_ty], [IntrNoMem]>; + def int_x86_sse2_cvtpd2dq : GCCBuiltin<"__builtin_ia32_cvtpd2dq">, + Intrinsic<[llvm_v4i32_ty, llvm_v2f64_ty], [IntrNoMem]>; + def int_x86_sse2_cvttpd2dq : GCCBuiltin<"__builtin_ia32_cvttpd2dq">, + Intrinsic<[llvm_v4i32_ty, llvm_v2f64_ty], [IntrNoMem]>; + def int_x86_sse2_cvtpd2ps : GCCBuiltin<"__builtin_ia32_cvtpd2ps">, + Intrinsic<[llvm_v4f32_ty, llvm_v2f64_ty], [IntrNoMem]>; + def int_x86_sse2_cvtps2dq : GCCBuiltin<"__builtin_ia32_cvtps2dq">, + Intrinsic<[llvm_v4i32_ty, llvm_v4f32_ty], [IntrNoMem]>; + def int_x86_sse2_cvttps2dq : GCCBuiltin<"__builtin_ia32_cvttps2dq">, + Intrinsic<[llvm_v4i32_ty, llvm_v4f32_ty], [IntrNoMem]>; + def int_x86_sse2_cvtps2pd : GCCBuiltin<"__builtin_ia32_cvtps2pd">, + Intrinsic<[llvm_v2f64_ty, llvm_v4f32_ty], [IntrNoMem]>; + def int_x86_sse2_cvtsd2si : GCCBuiltin<"__builtin_ia32_cvtsd2si">, + Intrinsic<[llvm_int_ty, llvm_v2f64_ty], [IntrNoMem]>; + def int_x86_sse2_cvttsd2si : GCCBuiltin<"__builtin_ia32_cvttsd2si">, + Intrinsic<[llvm_int_ty, llvm_v2f64_ty], [IntrNoMem]>; +} + // SIMD load ops let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.". def int_x86_sse2_loadu_pd : GCCBuiltin<"__builtin_ia32_loadupd">, diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td index 65e8aaeffa5..e0baec11e64 100644 --- a/lib/Target/X86/X86InstrSSE.td +++ b/lib/Target/X86/X86InstrSSE.td @@ -524,6 +524,23 @@ def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src), [(set FR64:$dst, (fextend (loadf32 addr:$src)))]>, XS, Requires<[HasSSE2]>; +// Aliases to match intrinsics which expect XMM operand(s). +def Int_CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops R32:$dst, VR128:$src), + "cvttsd2si {$src, $dst|$dst, $src}", + [(set R32:$dst, (int_x86_sse2_cvttsd2si VR128:$src))]>; +def Int_CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops R32:$dst, f128mem:$src), + "cvttsd2si {$src, $dst|$dst, $src}", + [(set R32:$dst, (int_x86_sse2_cvttsd2si + (load addr:$src)))]>; + +def CVTSD2SIrr: SDI<0x2D, MRMSrcReg, (ops R32:$dst, VR128:$src), + "cvtsd2si {$src, $dst|$dst, $src}", + [(set R32:$dst, (int_x86_sse2_cvtsd2si VR128:$src))]>; +def CVTSD2SIrm: SDI<0x2D, MRMSrcMem, (ops R32:$dst, f128mem:$src), + "cvtsd2si {$src, $dst|$dst, $src}", + [(set R32:$dst, (int_x86_sse2_cvtsd2si + (load addr:$src)))]>; + // Comparison instructions let isTwoAddress = 1 in { def CMPSSrr : SSI<0xC2, MRMSrcReg, @@ -800,62 +817,100 @@ def MOVHLPSrr : PSI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2), } // Conversion instructions -def CVTPI2PSr : PSI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src), +def CVTPI2PSrr : PSI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src), "cvtpi2ps {$src, $dst|$dst, $src}", []>; -def CVTPI2PSm : PSI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src), +def CVTPI2PSrm : PSI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src), "cvtpi2ps {$src, $dst|$dst, $src}", []>; -def CVTPI2PDr : PDI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src), +def CVTPI2PDrr : PDI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src), "cvtpi2pd {$src, $dst|$dst, $src}", []>; -def CVTPI2PDm : PDI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src), +def CVTPI2PDrm : PDI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src), "cvtpi2pd {$src, $dst|$dst, $src}", []>; // SSE2 instructions without OpSize prefix -def CVTDQ2PSr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src), - "cvtdq2ps {$src, $dst|$dst, $src}", []>, TB, - Requires<[HasSSE2]>; -def CVTDQ2PSm : I<0x5B, MRMSrcMem, (ops VR128:$dst, i128mem:$src), - "cvtdq2ps {$src, $dst|$dst, $src}", []>, TB, - Requires<[HasSSE2]>; +def CVTDQ2PSrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtdq2ps {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>, + TB, Requires<[HasSSE2]>; +def CVTDQ2PSrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, i128mem:$src), + "cvtdq2ps {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtdq2ps + (load addr:$src)))]>, + TB, Requires<[HasSSE2]>; // SSE2 instructions with XS prefix -def CVTDQ2PDr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR64:$src), - "cvtdq2pd {$src, $dst|$dst, $src}", []>, - XS, Requires<[HasSSE2]>; -def CVTDQ2PDm : I<0xE6, MRMSrcMem, (ops VR128:$dst, i64mem:$src), - "cvtdq2pd {$src, $dst|$dst, $src}", []>, - XS, Requires<[HasSSE2]>; +def CVTDQ2PDrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtdq2pd {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>, + XS, Requires<[HasSSE2]>; +def CVTDQ2PDrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, i64mem:$src), + "cvtdq2pd {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtdq2pd + (load addr:$src)))]>, + XS, Requires<[HasSSE2]>; -def CVTPS2PIr : PSI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src), +def CVTPS2PIrr : PSI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src), "cvtps2pi {$src, $dst|$dst, $src}", []>; -def CVTPS2PIm : PSI<0x2D, MRMSrcMem, (ops VR64:$dst, f64mem:$src), +def CVTPS2PIrm : PSI<0x2D, MRMSrcMem, (ops VR64:$dst, f64mem:$src), "cvtps2pi {$src, $dst|$dst, $src}", []>; -def CVTPD2PIr : PDI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src), +def CVTPD2PIrr : PDI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src), "cvtpd2pi {$src, $dst|$dst, $src}", []>; -def CVTPD2PIm : PDI<0x2D, MRMSrcMem, (ops VR64:$dst, f128mem:$src), +def CVTPD2PIrm : PDI<0x2D, MRMSrcMem, (ops VR64:$dst, f128mem:$src), "cvtpd2pi {$src, $dst|$dst, $src}", []>; -def CVTPS2DQr : PDI<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src), - "cvtps2dq {$src, $dst|$dst, $src}", []>; -def CVTPS2DQm : PDI<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src), - "cvtps2dq {$src, $dst|$dst, $src}", []>; +def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtps2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>; +def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src), + "cvtps2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtps2dq + (load addr:$src)))]>; +// SSE2 packed instructions with XS prefix +def CVTTPS2DQrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvttps2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))]>, + XS, Requires<[HasSSE2]>; +def CVTTPS2DQrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src), + "cvttps2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvttps2dq + (load addr:$src)))]>, + XS, Requires<[HasSSE2]>; + // SSE2 packed instructions with XD prefix -def CVTPD2DQr : SDI<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src), - "cvtpd2dq {$src, $dst|$dst, $src}", []>; -def CVTPD2DQm : SDI<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src), - "cvtpd2dq {$src, $dst|$dst, $src}", []>; +def CVTPD2DQrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtpd2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>, + XD, Requires<[HasSSE2]>; +def CVTPD2DQrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src), + "cvtpd2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtpd2dq + (load addr:$src)))]>, + XD, Requires<[HasSSE2]>; +def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvttpd2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>; +def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src), + "cvttpd2dq {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvttpd2dq + (load addr:$src)))]>; // SSE2 instructions without OpSize prefix -def CVTPS2PDr : I<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src), - "cvtps2pd {$src, $dst|$dst, $src}", []>, TB, - Requires<[HasSSE2]>; -def CVTPS2PDm : I<0x5A, MRMSrcReg, (ops VR128:$dst, f64mem:$src), - "cvtps2pd {$src, $dst|$dst, $src}", []>, TB, - Requires<[HasSSE2]>; +def CVTPS2PDrr : I<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtps2pd {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>, + TB, Requires<[HasSSE2]>; +def CVTPS2PDrm : I<0x5A, MRMSrcReg, (ops VR128:$dst, f64mem:$src), + "cvtps2pd {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtps2pd + (load addr:$src)))]>, + TB, Requires<[HasSSE2]>; -def CVTPD2PSr : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src), - "cvtpd2ps {$src, $dst|$dst, $src}", []>; -def CVTPD2PSm : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, f128mem:$src), - "cvtpd2ps {$src, $dst|$dst, $src}", []>; +def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src), + "cvtpd2ps {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>; +def CVTPD2PSrm : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, f128mem:$src), + "cvtpd2ps {$src, $dst|$dst, $src}", + [(set VR128:$dst, (int_x86_sse2_cvtpd2ps + (load addr:$src)))]>; // Arithmetic let isTwoAddress = 1 in {