gallivm: use nearest rounding for float->unorm24 conversion

Previously we were using truncation, which gives the correct result
only for numbers in [0.5-1.0] range (because there's no mantissa bits
to do any rounding there).
This is frequently hit (and probably only used there) when converting
fragment depth to depth format (d24s8 etc.) or otherwise dealing with
depth format.

v2: as spotted by Jose, get rid of extra type (src_type is already unsigned).

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.c b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
index ba51ff794f657da5b15d568b2f385120238f197a..712ce5f92dca1607db2d8db2a7ca88961a72f5f5 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_conv.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
@@ -266,17 +266,19 @@ lp_build_clamped_float_to_unsigned_norm(struct gallivm_state *gallivm,
    else if (dst_width == (mantissa + 1)) {
       /*
        * The destination width matches exactly what can be represented in
-       * floating point (i.e., mantissa + 1 bits). So do a straight
-       * multiplication followed by casting. No further rounding is necessary.
+       * floating point (i.e., mantissa + 1 bits). Even so correct rounding
+       * still needs to be applied (only for numbers in [0.5-1.0] would
+       * conversion using truncation after scaling be sufficient).
        */
-
       double scale;
+      struct lp_build_context uf32_bld;
 
+      lp_build_context_init(&uf32_bld, gallivm, src_type);
       scale = (double)((1ULL << dst_width) - 1);
 
       res = LLVMBuildFMul(builder, src,
                           lp_build_const_vec(gallivm, src_type, scale), "");
-      res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
+      res = lp_build_iround(&uf32_bld, res);
    }
    else {
       /*