[LLVMbugs] [Bug 6960] New: x86 codegen: vectors don't work on non-SSE CPUs

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=6960

           Summary: x86 codegen: vectors don't work on non-SSE CPUs
           Product: libraries
           Version: 2.7
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: edwintorok at gmail.com
                CC: llvmbugs at cs.uiuc.edu


Created an attachment (id=4751)
 --> (http://llvm.org/bugs/attachment.cgi?id=4751)
bitcode

Reported on #dri-devel, that softpipe/llvmpipe in mesa crashes like this, on a
CPU without SSE or SSE2:
#
glxgears: X86ISelLowering.cpp:6058: llvm::SDValue
llvm::X86TargetLowering::LowerVSETCC(llvm::SDValue, llvm::SelectionDAG&):
Assertion `VT0 == MVT::v4f32 || VT0 == MVT::v2f64' failed.
#
Stack dump:
#
0.      Running pass 'X86 DAG->DAG Instruction Selection' on function
'@draw_llvm_shader'

I dumped the module generated by mesa, and attached it here.

You can reproduce the bug like this (on llvm 2.7):
$ llc l.bc -march=x86 -mattr=-sse
UNREACHABLE executed!
0  llc 0x08631db9
Stack dump:
0.      Program arguments: /home/edwin/llvm-git/obj32/Release-Asserts/bin/llc
l.bc -march=x86 -mattr=-sse
1.      Running pass 'X86 DAG->DAG Instruction Selection' on function
'@draw_llvm_shader'
Aborted

And like this:
$ llc l.bc -march=x86 -mattr=-sse2
LLVM ERROR: Cannot yet select: 0x9d86fe4: v2i32 = X86ISD::CMPPD 0x9d98964,
0x9d7eccc, 0x9d7308c [ID=518]
  0x9d98964: v2f32 = bit_convert 0x9d988e0 [ID=512]
    0x9d988e0: v1i64,ch = load 0x9d98754, 0x9d987d8, 0x9d7d344<LD8[<unknown>]>
[ID=506]
      0x9d98754: ch = store 0x9d54080, 0x9d88918, 0x9d986d0, 
.....

/proc/cpuinfo:
#
processor       : 0
#
vendor_id       : AuthenticAMD
#
cpu family      : 6
#
model           : 3
#
model name      : AMD Duron(tm) Processor
#
stepping        : 1
#
cpu MHz         : 950.024
#
cache size      : 64 KB
#
fdiv_bug        : no
#
hlt_bug         : no
#
f00f_bug        : no
#
coma_bug        : no
#
fpu             : yes
#
fpu_exception   : yes
#
cpuid level     : 1
#
wp              : yes
#
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow up
#
bogomips        : 1900.04
#
clflush size    : 32
#
cache_alignment : 32
#
address sizes   : 36 bits physical, 32 bits virtual
#
power management:

For reference here is how I dumped the module:
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c
b/src/gallium/auxiliary/draw/draw_llvm.c
index 2738322..8e0a939 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -179,6 +179,7 @@ draw_llvm_create(struct draw_context *draw)

    llvm->pass = LLVMCreateFunctionPassManager(llvm->provider);
    LLVMAddTargetData(llvm->target, llvm->pass);
+   fprintf(stderr, "%s\n", LLVMCopyStringRepOfTargetData(llvm->target));
    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
     * but there are more on SVN. */
    /* TODO: Add more passes */
@@ -694,6 +695,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct
draw_llvm_variant *variant)
       debug_printf("\n");
    }
    variant->jit_func =
(draw_jit_vert_func)LLVMGetPointerToGlobal(llvm->draw->engine,
variant->function);
+   LLVMDumpModule(llvm->module);

    if (0)
       lp_disassemble(variant->jit_func);

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.