llvmpipe: use 64bit counter for occlusion queries

Some APIs require 64bit and at least for 64bit archs the overhead
should be minimal.

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

diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
index edb59cc8db6ea9b0e480822cd742603667b99147..79891cf5d832f47823eaf9ce659ac8e7ade42fb2 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
@@ -429,7 +429,7 @@ get_s_shift_and_mask(const struct util_format_description *format_desc,
  * Test the depth mask. Add the number of channel which has none zero mask
  * into the occlusion counter. e.g. maskvalue is {-1, -1, -1, -1}.
  * The counter will add 4.
- * TODO: could get that out of the loop, and need to use 64bit counter.
+ * TODO: could get that out of the fs loop.
  *
  * \param type holds element type of the mask vector.
  * \param maskvalue is the depth test mask.
@@ -458,6 +458,7 @@ lp_build_occlusion_count(struct gallivm_state *gallivm,
                                       LLVMInt32TypeInContext(context), bits);
       count = lp_build_intrinsic_unary(builder, popcntintr,
                                        LLVMInt32TypeInContext(context), bits);
+      count = LLVMBuildZExt(builder, count, LLVMIntTypeInContext(context, 64), "");
    }
    else if(util_cpu_caps.has_avx && type.length == 8) {
       const char *movmskintr = "llvm.x86.avx.movmsk.ps.256";
@@ -468,6 +469,7 @@ lp_build_occlusion_count(struct gallivm_state *gallivm,
                                       LLVMInt32TypeInContext(context), bits);
       count = lp_build_intrinsic_unary(builder, popcntintr,
                                        LLVMInt32TypeInContext(context), bits);
+      count = LLVMBuildZExt(builder, count, LLVMIntTypeInContext(context, 64), "");
    }
    else {
       unsigned i;
@@ -510,8 +512,11 @@ lp_build_occlusion_count(struct gallivm_state *gallivm,
        }
        count = lp_build_intrinsic_unary(builder, popcntintr, counttype, countd);
 
-       if (type.length > 4) {
-          count = LLVMBuildTrunc(builder, count, LLVMIntTypeInContext(context, 32), "");
+       if (type.length > 8) {
+          count = LLVMBuildTrunc(builder, count, LLVMIntTypeInContext(context, 64), "");
+       }
+       else if (type.length < 8) {
+          count = LLVMBuildZExt(builder, count, LLVMIntTypeInContext(context, 64), "");
        }
    }
    newcount = LLVMBuildLoad(builder, counter, "origcount");

diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index f517b67fff6d6a74c7a1cc1262b02e0a4f718af0..fa0f128ee3fd04ee65f7385d69d7192e1f92f6e7 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -195,7 +195,7 @@ lp_jit_create_types(struct lp_fragment_shader_variant *lp)
       LLVMTypeRef elem_types[LP_JIT_THREAD_DATA_COUNT];
       LLVMTypeRef thread_data_type;
 
-      elem_types[LP_JIT_THREAD_DATA_COUNTER] = LLVMInt32TypeInContext(lc);
+      elem_types[LP_JIT_THREAD_DATA_COUNTER] = LLVMInt64TypeInContext(lc);
 
       thread_data_type = LLVMStructTypeInContext(lc, elem_types,
                                                  Elements(elem_types), 0);

diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 2ecfde75a09cd1a4e9b7fcb9c7e2b90c789682fe..30cfaaef5d6afb827b356ca352aa973112147b5a 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -164,7 +164,7 @@ enum {
 
 struct lp_jit_thread_data
 {
-   uint32_t vis_counter;
+   uint64_t vis_counter;
 };