1 files changed, 43 insertions, 12 deletions

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
index 7c585fd..f3df3dd 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
@@ -42,6 +42,26 @@
 #include "lp_bld_logic.h"
 
 
+/*
+ * XXX
+ *
+ * Selection with vector conditional like
+ *
+ *    select <4 x i1> %C, %A, %B
+ *
+ * is valid IR (e.g. llvm/test/Assembler/vector-select.ll), but it is not
+ * supported on any backend.
+ *
+ * Expanding the boolean vector to full SIMD register width, as in
+ *
+ *    sext <4 x i1> %C to <4 x i32>
+ *
+ * is valid and supported (e.g., llvm/test/CodeGen/X86/vec_compare.ll), but
+ * it causes assertion failures in LLVM 2.6. It appears to work correctly on 
+ * LLVM 2.7.
+ */
+
+
 /**
  * Build code to compare two values 'a' and 'b' of 'type' using the given func.
  * \param func  one of PIPE_FUNC_x
@@ -54,13 +74,11 @@ lp_build_compare(LLVMBuilderRef builder,
                  LLVMValueRef a,
                  LLVMValueRef b)
 {
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
    LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
    LLVMValueRef zeros = LLVMConstNull(int_vec_type);
    LLVMValueRef ones = LLVMConstAllOnes(int_vec_type);
    LLVMValueRef cond;
    LLVMValueRef res;
-   unsigned i;
 
    assert(func >= PIPE_FUNC_NEVER);
    assert(func <= PIPE_FUNC_ALWAYS);
@@ -74,10 +92,12 @@ lp_build_compare(LLVMBuilderRef builder,
 
    /* XXX: It is not clear if we should use the ordered or unordered operators */
 
+#if HAVE_LLVM < 0x0207
 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
    if(type.width * type.length == 128) {
       if(type.floating && util_cpu_caps.has_sse) {
          /* float[4] comparison */
+         LLVMTypeRef vec_type = lp_build_vec_type(type);
          LLVMValueRef args[3];
          unsigned cc;
          boolean swap;
@@ -147,6 +167,7 @@ lp_build_compare(LLVMBuilderRef builder,
          const char *pcmpgt;
          LLVMValueRef args[2];
          LLVMValueRef res;
+         LLVMTypeRef vec_type = lp_build_vec_type(type);
 
          switch (type.width) {
          case 8:
@@ -200,6 +221,7 @@ lp_build_compare(LLVMBuilderRef builder,
       }
    } /* if (type.width * type.length == 128) */
 #endif
+#endif /* HAVE_LLVM < 0x0207 */
 
    if(type.floating) {
       LLVMRealPredicate op;
@@ -233,16 +255,19 @@ lp_build_compare(LLVMBuilderRef builder,
          return lp_build_undef(type);
       }
 
-#if 0
-      /* XXX: Although valid IR, no LLVM target currently support this */
+#if HAVE_LLVM >= 0x0207
       cond = LLVMBuildFCmp(builder, op, a, b, "");
-      res = LLVMBuildSelect(builder, cond, ones, zeros, "");
+      res = LLVMBuildSExt(builder, cond, int_vec_type, "");
 #else
-      res = LLVMGetUndef(int_vec_type);
       if (type.length == 1) {
-         res = LLVMBuildFCmp(builder, op, a, b, "");
+         cond = LLVMBuildFCmp(builder, op, a, b, "");
+         res = LLVMBuildSExt(builder, cond, int_vec_type, "");
       }
       else {
+         unsigned i;
+
+         res = LLVMGetUndef(int_vec_type);
+
          debug_printf("%s: warning: using slow element-wise float"
                       " vector comparison\n", __FUNCTION__);
          for (i = 0; i < type.length; ++i) {
@@ -286,16 +311,19 @@ lp_build_compare(LLVMBuilderRef builder,
          return lp_build_undef(type);
       }
 
-#if 0
-      /* XXX: Although valid IR, no LLVM target currently support this */
+#if HAVE_LLVM >= 0x0207
       cond = LLVMBuildICmp(builder, op, a, b, "");
-      res = LLVMBuildSelect(builder, cond, ones, zeros, "");
+      res = LLVMBuildSExt(builder, cond, int_vec_type, "");
 #else
-      res = LLVMGetUndef(int_vec_type);
       if (type.length == 1) {
-         res = LLVMBuildICmp(builder, op, a, b, "");
+         cond = LLVMBuildICmp(builder, op, a, b, "");
+         res = LLVMBuildSExt(builder, cond, int_vec_type, "");
       }
       else {
+         unsigned i;
+
+         res = LLVMGetUndef(int_vec_type);
+
          debug_printf("%s: warning: using slow element-wise int"
                       " vector comparison\n", __FUNCTION__);
 
@@ -337,6 +365,8 @@ lp_build_cmp(struct lp_build_context *bld,
 
 /**
  * Return mask ? a : b;
+ *
+ * mask is a bitwise mask, composed of 0 or ~0 for each element.
  */
 LLVMValueRef
 lp_build_select(struct lp_build_context *bld,
@@ -351,6 +381,7 @@ lp_build_select(struct lp_build_context *bld,
       return a;
 
    if (type.length == 1) {
+      mask = LLVMBuildTrunc(bld->builder, mask, LLVMInt1Type(), "");
       res = LLVMBuildSelect(bld->builder, mask, a, b, "");
    }
    else {