gallium/swr: fix corruptions in Unigine Heaven

Few changes to fix the last corruptions in Heaven:
- fix indirect TCS input when vertex/attribute
  index is not the same for each patch
- use the correct functions to build loops in shader
- fix using vmask for writting TCS output

Reviewed-by: Krzysztof Raszkowski <krzysztof.raszkowski@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3980>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3980>

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_ir_common.h b/src/gallium/auxiliary/gallivm/lp_bld_ir_common.h
index 34c09193936c84b5a2e0f89d10580d4c8c2bdc64..ebcf9a21b2635f5f85fbf1ed5df90f7189af3f1a 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_ir_common.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_ir_common.h
@@ -29,6 +29,10 @@
 #ifndef LP_BLD_IR_COMMON_H
 #define LP_BLD_IR_COMMON_H
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
 #include "gallivm/lp_bld.h"
 #include "gallivm/lp_bld_limits.h"
 
@@ -117,4 +121,8 @@ void lp_exec_continue(struct lp_exec_mask *mask);
 
 void lp_exec_break(struct lp_exec_mask *mask, int *pc, bool break_always);
 
+#ifdef __cplusplus
+}
+#endif
+
 #endif

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
index 00fb0268dd60d2a9308e944d0c99eee21b159b53..b8bbfb2bf13401c965e829ae32cb6ae981d7766e 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
@@ -36,6 +36,9 @@
 #ifndef LP_BLD_LOGIC_H
 #define LP_BLD_LOGIC_H
 
+#ifdef __cplusplus
+extern "C" {
+#endif
 
 #include "gallivm/lp_bld.h"
 
@@ -94,4 +97,8 @@ lp_build_any_true_range(struct lp_build_context *bld,
                         unsigned real_length,
                         LLVMValueRef val);
 
+#ifdef __cplusplus
+}
+#endif
+
 #endif /* !LP_BLD_LOGIC_H */

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 2586830bb468fcfbf9912c7887133d264bb62843..0ab1aa1fe0212d228c7bd86f2f8750afe9b44120 100644 (file)
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -1764,7 +1764,8 @@ emit_store_tcs_output(struct lp_build_tgsi_context *bld_base,
                                           reg->Register.Indirect,
                                           attrib_index,
                                           channel_index,
-                                          value, NULL);
+                                          value,
+                                          mask_vec(bld_base));
 }
 
 static void

diff --git a/src/gallium/drivers/swr/swr_shader.cpp b/src/gallium/drivers/swr/swr_shader.cpp
index 943215703698326f19767b7e0d96b096fc3b90e1..90333536421f0e7de68836a6014be1ad573081b5 100644 (file)
--- a/src/gallium/drivers/swr/swr_shader.cpp
+++ b/src/gallium/drivers/swr/swr_shader.cpp
@@ -52,6 +52,7 @@
 #include "gallivm/lp_bld_tgsi.h"
 #include "gallivm/lp_bld_const.h"
 #include "gallivm/lp_bld_printf.h"
+#include "gallivm/lp_bld_logic.h"
 
 #include "swr_context.h"
 #include "gen_surf_state_llvm.h"
@@ -72,10 +73,18 @@
 
 #include "gallivm/lp_bld_type.h"
 
-#ifdef DEBUG
-constexpr bool verbose_shader = true;
+#if defined(DEBUG) && defined(SWR_VERBOSE_SHADER)
+constexpr bool verbose_shader          = true;
+constexpr bool verbose_tcs_shader_in   = true;
+constexpr bool verbose_tcs_shader_out  = true;
+constexpr bool verbose_tcs_shader_loop = true;
+constexpr bool verbose_vs_shader       = true;
 #else
-constexpr bool verbose_shader = false;
+constexpr bool verbose_shader          = false;
+constexpr bool verbose_tcs_shader_in   = false;
+constexpr bool verbose_tcs_shader_out  = false;
+constexpr bool verbose_tcs_shader_loop = false;
+constexpr bool verbose_vs_shader       = false;
 #endif
 
 using namespace SwrJit;
@@ -397,7 +406,8 @@ struct BuilderSWR : public Builder {
                             boolean is_aindex_indirect,
                             LLVMValueRef attrib_index,
                             LLVMValueRef swizzle_index,
-                            LLVMValueRef value);
+                            LLVMValueRef value,
+                            LLVMValueRef mask_vec);
 
    // Barrier implementation (available only in TCS)
    void
@@ -447,7 +457,7 @@ struct swr_tcs_llvm_iface {
 
    uint32_t output_vertices;
 
-   struct lp_build_for_loop_state loop_state;
+   LLVMValueRef loop_var;
 
    Value *pVtxAttribMap;
    Value *pVtxOutputAttribMap;
@@ -615,7 +625,8 @@ void swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface,
                                                is_aindex_indirect,
                                                attrib_index,
                                                swizzle_index,
-                                               value);
+                                               value,
+                                               mask_vec);
 }
 
 
@@ -931,20 +942,20 @@ BuilderSWR::swr_tcs_llvm_emit_prologue(struct lp_build_tgsi_soa_context* bld)
 {
    swr_tcs_llvm_iface *iface = (swr_tcs_llvm_iface*)bld->tcs_iface;
 
-   // Iterate for all the vertices in the output patch
-   lp_build_for_loop_begin(&iface->loop_state, gallivm,
-                        lp_build_const_int32(gallivm, 0),
-                        LLVMIntULT,
-                        lp_build_const_int32(gallivm, iface->output_vertices),
-                        lp_build_const_int32(gallivm, 1));
+   Value* loop_var = ALLOCA(mSimdInt32Ty);
+   STORE(VBROADCAST(C(0)), loop_var);
+
+   iface->loop_var = wrap(loop_var);
+
+   lp_exec_bgnloop(&bld->exec_mask, true);
 
    IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
-   bld->system_values.invocation_id  = wrap(VBROADCAST(unwrap(iface->loop_state.counter)));
+   bld->system_values.invocation_id = wrap((LOAD(unwrap(iface->loop_var))));
 
-   if (verbose_shader) {
-      lp_build_printf(gallivm, "Prologue LOOP: Iteration %d BEGIN\n", iface->loop_state.counter);
-      lp_build_print_value(gallivm, "LOOP: InvocationId: \n", bld->system_values.invocation_id);
+   if (verbose_tcs_shader_loop) {
+      lp_build_print_value(gallivm, "Prologue LOOP Iteration BEGIN:", bld->system_values.invocation_id);
    }
+
 }
 
 void
@@ -952,10 +963,19 @@ BuilderSWR::swr_tcs_llvm_emit_epilogue(struct lp_build_tgsi_soa_context* bld)
 {
    swr_tcs_llvm_iface *iface = (swr_tcs_llvm_iface*)bld->tcs_iface;
 
-   if (verbose_shader) {
-      lp_build_printf(gallivm, "Epilogue LOOP: Iteration %d END\n", iface->loop_state.counter);
+   struct lp_build_context *uint_bld = &bld->bld_base.uint_bld;
+
+   STORE(ADD(LOAD(unwrap(iface->loop_var)), VBROADCAST(C(1))), unwrap(iface->loop_var));
+   if (verbose_tcs_shader_loop) {
+      lp_build_print_value(gallivm, "Epilogue LOOP: ", wrap(LOAD(unwrap(iface->loop_var))));
    }
-   lp_build_for_loop_end(&iface->loop_state);
+
+   LLVMValueRef tmp = lp_build_cmp(uint_bld, PIPE_FUNC_GEQUAL, wrap(LOAD(unwrap(iface->loop_var))),
+                                   wrap(VBROADCAST(C(iface->output_vertices))));
+   lp_exec_mask_cond_push(&bld->exec_mask, tmp);
+   lp_exec_break(&bld->exec_mask, &bld->bld_base.pc, false);
+   lp_exec_mask_cond_pop(&bld->exec_mask);
+   lp_exec_endloop(bld->bld_base.base.gallivm, &bld->exec_mask);
 }
 
 LLVMValueRef
@@ -968,46 +988,75 @@ BuilderSWR::swr_tcs_llvm_fetch_input(const struct lp_build_tcs_iface *tcs_iface,
                                      LLVMValueRef swizzle_index)
 {
    swr_tcs_llvm_iface *iface = (swr_tcs_llvm_iface*)tcs_iface;
+
    Value *vert_index = unwrap(vertex_index);
    Value *attr_index = unwrap(attrib_index);
 
    IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
 
-   if (verbose_shader) {
-      lp_build_print_value(gallivm, "TCS: Vertex index: ", vertex_index);
-      lp_build_print_value(gallivm, "TCS: Attrib index: ", attrib_index);
-      lp_build_print_value(gallivm, "TCS: Swizzle index: ", swizzle_index);
+   if (verbose_tcs_shader_in) {
+      lp_build_printf(gallivm, "[TCS IN][VTX] ======================================\n");
+      lp_build_print_value(gallivm, "[TCS IN][VTX] vertex_index: ", vertex_index);
+      lp_build_print_value(gallivm, "[TCS IN][VTX] attrib_index: ", attrib_index);
+      lp_build_printf(gallivm, "[TCS IN][VTX] --------------------------------------\n");
    }
 
-   if (is_vindex_indirect) {
-      vert_index = VEXTRACT(vert_index, C(0));
-      if (verbose_shader) {
-         lp_build_print_value(gallivm, "TCS: Extracted vertex index: ", vertex_index);
-      }
-   }
+   Value *res = unwrap(bld_base->base.zero);
+   if (is_vindex_indirect || is_aindex_indirect) {
+      int i;
+      struct lp_type type = bld_base->base.type;
 
-   if (is_aindex_indirect) {
-      attr_index = VEXTRACT(attr_index, C(0));
-      if (verbose_shader) {
-         lp_build_print_value(gallivm, "TCS: Extracted attrib index: ", attrib_index);
-      }
-   }
+      for (i = 0; i < type.length; i++) {
+         Value *vert_chan_index = vert_index;
+         Value *attr_chan_index = attr_index;
 
-   Value *attrib = LOAD(GEP(iface->pVtxAttribMap, {C(0), attr_index}));
-   if (verbose_shader) {
-      lp_build_print_value(gallivm, "TCS: Attrib index loaded from map: ", wrap(attrib));
-   }
+         if (is_vindex_indirect) {
+            vert_chan_index = VEXTRACT(vert_index, C(i));
+         }
+         if (is_aindex_indirect) {
+            attr_chan_index = VEXTRACT(attr_index, C(i));
+         }
 
-   Value *pBase = GEP(iface->pTcsCtx,
-                     { C(0), C(SWR_HS_CONTEXT_vert), vert_index,
-                     C(simdvertex_attrib), attrib /*attr_index*/, unwrap(swizzle_index) });
+         Value *attrib =
+            LOAD(GEP(iface->pVtxAttribMap, {C(0), attr_chan_index}));
 
-   LLVMValueRef res = wrap(LOAD(pBase));
+         Value *pBase = GEP(iface->pTcsCtx,
+                        { C(0), C(SWR_HS_CONTEXT_vert), vert_chan_index,
+                        C(simdvertex_attrib), attrib, unwrap(swizzle_index), C(i) });
 
-   if (verbose_shader) {
-      lp_build_print_value(gallivm, "TCS input fetched: ", res);
+         Value *val = LOAD(pBase);
+
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS IN][VTX] vert_chan_index: ", wrap(vert_chan_index));
+            lp_build_print_value(gallivm, "[TCS IN][VTX] attrib_index: ", attrib_index);
+            lp_build_print_value(gallivm, "[TCS IN][VTX] attr_chan_index: ", wrap(attr_index));
+            lp_build_print_value(gallivm, "[TCS IN][VTX] attrib read from map: ", wrap(attrib));
+            lp_build_print_value(gallivm, "[TCS IN][VTX] swizzle_index: ", swizzle_index);
+            lp_build_print_value(gallivm, "[TCS IN][VTX] Loaded: ", wrap(val));
+         }
+         res = VINSERT(res, val, C(i));
+      }
+   } else {
+      Value *attrib = LOAD(GEP(iface->pVtxAttribMap, {C(0), attr_index}));
+
+      Value *pBase = GEP(iface->pTcsCtx,
+                        { C(0), C(SWR_HS_CONTEXT_vert), vert_index,
+                        C(simdvertex_attrib), attrib, unwrap(swizzle_index) });
+
+      res = LOAD(pBase);
+
+      if (verbose_tcs_shader_in) {
+         lp_build_print_value(gallivm, "[TCS IN][VTX] attrib_index: ", attrib_index);
+         lp_build_print_value(gallivm, "[TCS IN][VTX] attr_chan_index: ", wrap(attr_index));
+         lp_build_print_value(gallivm, "[TCS IN][VTX] attrib read from map: ", wrap(attrib));
+         lp_build_print_value(gallivm, "[TCS IN][VTX] swizzle_index: ", swizzle_index);
+         lp_build_print_value(gallivm, "[TCS IN][VTX] Loaded: ", wrap(res));
+      }
+   }
+   if (verbose_tcs_shader_in) {
+      lp_build_print_value(gallivm, "[TCS IN][VTX] returning: ", wrap(res));
    }
-   return res;
+   return wrap(res);
 }
 
 LLVMValueRef
@@ -1027,25 +1076,10 @@ BuilderSWR::swr_tcs_llvm_fetch_output(const struct lp_build_tcs_iface *tcs_iface
 
    IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
 
-   if (verbose_shader) {
-      lp_build_print_value(gallivm, "++TCSo: Vertex index: ", vertex_index);
-      lp_build_print_value(gallivm, "++TCSo: Attrib index: ", wrap(attr_index));
-      lp_build_print_value(gallivm, "++TCSo: Swizzle index: ", swizzle_index);
-   }
-
-   if (is_vindex_indirect) {
-      vert_index = VEXTRACT(vert_index, C(0));
-      if (verbose_shader)
-      {
-         lp_build_print_value(gallivm, "TCSo: Extracted vertex index: ", vertex_index);
-      }
-   }
-
-   if (is_aindex_indirect) {
-      attr_index = VEXTRACT(attr_index, C(0));
-      if (verbose_shader) {
-         lp_build_print_value(gallivm, "TCSo: Extracted attrib index: ", attrib_index);
-      }
+   if (verbose_tcs_shader_in) {
+      lp_build_print_value(gallivm, "[TCS INOUT] Vertex index: ", vertex_index);
+      lp_build_print_value(gallivm, "[TCS INOUT] Attrib index: ", wrap(attr_index));
+      lp_build_print_value(gallivm, "[TCS INOUT] Swizzle index: ", swizzle_index);
    }
 
    Value* res = unwrap(bld_base->base.zero);
@@ -1054,8 +1088,24 @@ BuilderSWR::swr_tcs_llvm_fetch_output(const struct lp_build_tcs_iface *tcs_iface
       Value* p1 = LOAD(iface->pTcsCtx, {0, SWR_HS_CONTEXT_pCPout});
       Value* pCpOut = GEP(p1, {lane});
 
-      if (name == TGSI_SEMANTIC_TESSOUTER || name == TGSI_SEMANTIC_TESSINNER) {
+      Value *vert_chan_index = vert_index;
+      Value *attr_chan_index = attr_index;
+
+      if (is_vindex_indirect) {
+         vert_chan_index = VEXTRACT(vert_index, C(lane));
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT] Extracted vertex index: ", wrap(vert_chan_index));
+         }
+      }
 
+      if (is_aindex_indirect) {
+         attr_chan_index = VEXTRACT(attr_index, C(lane));
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT] Extracted attrib index: ", wrap(attr_chan_index));
+         }
+      }
+
+      if (name == TGSI_SEMANTIC_TESSOUTER || name == TGSI_SEMANTIC_TESSINNER) {
          Value* tessFactors = GEP(pCpOut, {(uint32_t)0, ScalarPatch_tessFactors});
          Value* tessFactorArray = nullptr;
          if (name == TGSI_SEMANTIC_TESSOUTER) {
@@ -1065,33 +1115,36 @@ BuilderSWR::swr_tcs_llvm_fetch_output(const struct lp_build_tcs_iface *tcs_iface
          }
          Value* tessFactor = GEP(tessFactorArray, {C(0), unwrap(swizzle_index)});
          res = VINSERT(res, LOAD(tessFactor), C(lane));
-
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT][FACTOR] lane (patch-id): ", wrap(C(lane)));
+            lp_build_print_value(gallivm, "[TCS INOUT][FACTOR] loaded value: ", wrap(res));
+         }
       } else if (name == TGSI_SEMANTIC_PATCH) {
-         lp_build_print_value(gallivm, "bbbbb TCS per-patch attr_index: ", wrap(attr_index));
-         Value* attr = GEP(pCpOut, {C(0), C(ScalarPatch_patchData), C(ScalarCPoint_attrib), attr_index, unwrap(swizzle_index)});
-         res = VINSERT(res, LOAD(attr), C(lane));
-         if (verbose_shader) {
-            lp_build_print_value(gallivm, "++TCSo per-patch lane (patch-id): ", wrap(C(lane)));
-            lp_build_print_value(gallivm, "++TCSo per-patch loaded value: ", wrap(res));
+         Value* attr_index_from_map = LOAD(GEP(iface->pPatchOutputAttribMap, {C(0), attr_chan_index}));
+         Value* attr_value = GEP(pCpOut, {C(0), C(ScalarPatch_patchData), C(ScalarCPoint_attrib), attr_index_from_map, unwrap(swizzle_index)});
+         res = VINSERT(res, LOAD(attr_value), C(lane));
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT][PATCH] attr index loaded from map: ", wrap(attr_index_from_map));
+            lp_build_print_value(gallivm, "[TCS INOUT][PATCH] lane (patch-id): ", wrap(C(lane)));
+            lp_build_print_value(gallivm, "[TCS INOUT][PATCH] loaded value: ", wrap(res));
          }
       } else {
          // Generic attribute
          Value *attrib =
-             LOAD(GEP(iface->pVtxOutputAttribMap, {C(0), attr_index}));
-         if (verbose_shader)
-         {
-            lp_build_print_value(gallivm, "TCSo: Attrib index from map: ", wrap(attrib));
+             LOAD(GEP(iface->pVtxOutputAttribMap, {C(0), attr_chan_index}));
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT][VTX] Attrib index from map: ", wrap(attrib));
          }
-         Value* attr_chan = GEP(pCpOut, {C(0), C(ScalarPatch_cp), vert_index,
+         Value* attr_chan = GEP(pCpOut, {C(0), C(ScalarPatch_cp), vert_chan_index,
                                     C(ScalarCPoint_attrib), attrib, unwrap(swizzle_index)});
 
          res = VINSERT(res, LOAD(attr_chan), C(lane));
+         if (verbose_tcs_shader_in) {
+            lp_build_print_value(gallivm, "[TCS INOUT][VTX] loaded value: ", wrap(res));
+         }
       }
    }
 
-   if (verbose_shader) {
-      lp_build_print_value(gallivm, "TCSo: output fetched: ", wrap(res));
-   }
    return wrap(res);
 }
 
@@ -1104,18 +1157,19 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
                                       boolean is_aindex_indirect,
                                       LLVMValueRef attrib_index,
                                       LLVMValueRef swizzle_index,
-                                      LLVMValueRef value)
+                                      LLVMValueRef value,
+                                      LLVMValueRef mask_vec)
 {
    swr_tcs_llvm_iface *iface = (swr_tcs_llvm_iface*)tcs_iface;
    struct lp_build_tgsi_soa_context* bld = (struct lp_build_tgsi_soa_context*)bld_base;
 
    IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
 
-    if (verbose_shader) {
+    if (verbose_tcs_shader_out) {
       lp_build_printf(gallivm, "[TCS OUT] =============================================\n");
     }
 
-   if (verbose_shader) {
+   if (verbose_tcs_shader_out) {
       lp_build_print_value(gallivm, "[TCS OUT] Store mask: ", bld->exec_mask.exec_mask);
       lp_build_print_value(gallivm, "[TCS OUT] Store value: ", value);
    }
@@ -1123,9 +1177,7 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
    Value *vert_index = unwrap(vertex_index);
    Value *attr_index = unwrap(attrib_index);
 
-   IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
-
-   if (verbose_shader) {
+   if (verbose_tcs_shader_out) {
       lp_build_print_value(gallivm, "[TCS OUT] Vertex index: ", vertex_index);
       lp_build_print_value(gallivm, "[TCS OUT] Attrib index: ", wrap(attr_index));
       lp_build_print_value(gallivm, "[TCS OUT] Swizzle index: ", swizzle_index);
@@ -1133,19 +1185,26 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
 
    if (is_vindex_indirect) {
       vert_index = VEXTRACT(vert_index, C(0));
-      if (verbose_shader)
-      {
+      if (verbose_tcs_shader_out) {
          lp_build_print_value(gallivm, "[TCS OUT] Extracted vertex index: ", vertex_index);
       }
    }
 
    if (is_aindex_indirect) {
       attr_index = VEXTRACT(attr_index, C(0));
-      if (verbose_shader) {
+      if (verbose_tcs_shader_out) {
          lp_build_print_value(gallivm, "[TCS OUT] Extracted attrib index: ", wrap(attr_index));
       }
    }
 
+   if (verbose_tcs_shader_out) {
+      if (bld->exec_mask.has_mask) {
+         lp_build_print_value(gallivm, "[TCS OUT] Exec mask: ", bld->exec_mask.exec_mask);
+      }
+      else {
+         lp_build_printf(gallivm, "[TCS OUT] has no mask\n");
+      }
+   }
    for (uint32_t lane = 0; lane < mVWidth; lane++) {
       Value* p1 = LOAD(iface->pTcsCtx, {0, SWR_HS_CONTEXT_pCPout});
       Value* pCpOut = GEP(p1, {lane});
@@ -1160,19 +1219,21 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
          }
          Value* tessFactor = GEP(tessFactorArray, {C(0), unwrap(swizzle_index)});
          Value* valueToStore = VEXTRACT(unwrap(value), C(lane));
-         struct lp_exec_mask *mask = &bld->exec_mask;
-         if (mask->has_mask) {
+         valueToStore = BITCAST(valueToStore, mFP32Ty);
+         if (mask_vec) {
             Value *originalVal = LOAD(tessFactor);
-            Value *vMask = TRUNC(VEXTRACT(unwrap(mask->exec_mask), C(lane)), mInt1Ty);
+            Value *vMask = TRUNC(VEXTRACT(unwrap(mask_vec), C(lane)), mInt1Ty);
             valueToStore = SELECT(vMask, valueToStore, originalVal);
          }
          STORE(valueToStore, tessFactor);
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out)
+         {
+            lp_build_print_value(gallivm, "[TCS OUT][FACTOR] Mask_vec mask: ", mask_vec);
             lp_build_print_value(gallivm, "[TCS OUT][FACTOR] Stored value: ", wrap(valueToStore));
          }
       } else if (name == TGSI_SEMANTIC_PATCH) {
          Value* attrib = LOAD(GEP(iface->pPatchOutputAttribMap, {C(0), attr_index}));
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out) {
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] vert_index: ", wrap(vert_index));
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] attr_index: ", wrap(attr_index));
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] vert_index_indirect: ", wrap(C(is_vindex_indirect)));
@@ -1181,33 +1242,34 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
          }
          Value* attr = GEP(pCpOut, {C(0), C(ScalarPatch_patchData), C(ScalarCPoint_attrib), attrib});
          Value* value_to_store = VEXTRACT(unwrap(value), C(lane));
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out) {
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] lane (patch-id): ", wrap(C(lane)));
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] value to store: ", value);
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] per-patch value to store: ", wrap(value_to_store));
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] chan_index: ", swizzle_index);
          }
-         struct lp_exec_mask *mask = &bld->exec_mask;
-         if (mask->has_mask) {
+         value_to_store = BITCAST(value_to_store, mFP32Ty);
+         if (mask_vec) {
             Value *originalVal = LOADV(attr, {C(0), unwrap(swizzle_index)});
-            Value *vMask = TRUNC(VEXTRACT(unwrap(mask->exec_mask), C(lane)), mInt1Ty);
-            value_to_store = SELECT(vMask, BITCAST(value_to_store, mFP32Ty), originalVal);
-            if (verbose_shader) {
-               lp_build_print_value(gallivm, "[TCS OUT][PATCH] store mask: ", bld->exec_mask.exec_mask);
+            Value *vMask = TRUNC(VEXTRACT(unwrap(mask_vec), C(lane)), mInt1Ty);
+            value_to_store = SELECT(vMask, value_to_store, originalVal);
+            if (verbose_tcs_shader_out) {
+               lp_build_print_value(gallivm, "[TCS OUT][PATCH] store mask: ", mask_vec);
                lp_build_print_value(gallivm, "[TCS OUT][PATCH] loaded original value: ", wrap(originalVal));
                lp_build_print_value(gallivm, "[TCS OUT][PATCH] vMask: ", wrap(vMask));
                lp_build_print_value(gallivm, "[TCS OUT][PATCH] selected value to store: ", wrap(value_to_store));
             }
          }
          STOREV(value_to_store, attr, {C(0), unwrap(swizzle_index)});
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out) {
             lp_build_print_value(gallivm, "[TCS OUT][PATCH] stored value: ", wrap(value_to_store));
          }
       } else {
          Value* value_to_store = VEXTRACT(unwrap(value), C(lane));
          Value* attrib = LOAD(GEP(iface->pVtxOutputAttribMap, {C(0), attr_index}));
 
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out) {
+            lp_build_printf(gallivm, "[TCS OUT] Writting attribute\n");
             lp_build_print_value(gallivm, "[TCS OUT][VTX] invocation_id: ", bld->system_values.invocation_id);
             lp_build_print_value(gallivm, "[TCS OUT][VTX] attribIndex: ", wrap(attr_index));
             lp_build_print_value(gallivm, "[TCS OUT][VTX] attrib read from map: ", wrap(attrib));
@@ -1221,23 +1283,21 @@ BuilderSWR::swr_tcs_llvm_store_output(const struct lp_build_tcs_iface *tcs_iface
                                     C(ScalarCPoint_attrib), attrib, unwrap(swizzle_index)});
 
          // Mask output values if needed
-         struct lp_exec_mask *mask = &bld->exec_mask;
-         if (mask->has_mask) {
+         value_to_store = BITCAST(value_to_store, mFP32Ty);
+         if (mask_vec) {
             Value *originalVal = LOAD(attr_chan);
-            Value *vMask = TRUNC(VEXTRACT(unwrap(mask->exec_mask), C(lane)), mInt1Ty);
-            // convert input to float before trying to store
-            value_to_store = SELECT(vMask, BITCAST(value_to_store, mFP32Ty), originalVal);
+            Value *vMask = TRUNC(VEXTRACT(unwrap(mask_vec), C(lane)), mInt1Ty);
+            value_to_store = SELECT(vMask, value_to_store, originalVal);
          }
          STORE(value_to_store, attr_chan);
-         if (verbose_shader) {
+         if (verbose_tcs_shader_out) {
+            lp_build_print_value(gallivm, "[TCS OUT][VTX] Mask_vec mask: ", mask_vec);
             lp_build_print_value(gallivm, "[TCS OUT][VTX] stored: ", wrap(value_to_store));
          }
       }
    }
 }
 
-
-
 void
 BuilderSWR::swr_tcs_llvm_emit_barrier(const struct lp_build_tcs_iface *tcs_iface,
                                       struct lp_build_tgsi_context *bld_base)
@@ -1245,28 +1305,34 @@ BuilderSWR::swr_tcs_llvm_emit_barrier(const struct lp_build_tcs_iface *tcs_iface
    swr_tcs_llvm_iface *iface = (swr_tcs_llvm_iface*)tcs_iface;
    struct lp_build_tgsi_soa_context* bld = (struct lp_build_tgsi_soa_context*)bld_base;
 
-   if (verbose_shader) {
-      lp_build_printf(gallivm, "Barrier LOOP: Iteration %d END\n", iface->loop_state.counter);
+   if (verbose_tcs_shader_loop) {
+      lp_build_print_value(gallivm, "Barrier LOOP: Iteration %d END\n", iface->loop_var);
    }
 
-   // End previous loop
-   lp_build_for_loop_end(&iface->loop_state);
+   struct lp_build_context *uint_bld = &bld->bld_base.uint_bld;
+
+   STORE(ADD(LOAD(unwrap(iface->loop_var)), VBROADCAST(C(1))), unwrap(iface->loop_var));
 
-   // Start new one
-   lp_build_for_loop_begin(&iface->loop_state, gallivm,
-                        lp_build_const_int32(gallivm, 0),
-                        LLVMIntULT,
-                        lp_build_const_int32(gallivm, iface->output_vertices),
-                        lp_build_const_int32(gallivm, 1));
+   LLVMValueRef tmp = lp_build_cmp(uint_bld, PIPE_FUNC_GEQUAL, wrap(LOAD(unwrap(iface->loop_var))),
+                                   wrap(VBROADCAST(C(iface->output_vertices))));
 
+   lp_exec_mask_cond_push(&bld->exec_mask, tmp);
+   lp_exec_break(&bld->exec_mask, &bld->bld_base.pc, false);
+   lp_exec_mask_cond_pop(&bld->exec_mask);
+   lp_exec_endloop(bld->bld_base.base.gallivm, &bld->exec_mask);
 
    IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
 
-   bld->system_values.invocation_id  = wrap(VBROADCAST(unwrap(iface->loop_state.counter)));
+   STORE(VBROADCAST(C(0)), unwrap(iface->loop_var));
+   lp_exec_bgnloop(&bld->exec_mask, true);
 
-   if (verbose_shader) {
-      lp_build_printf(gallivm, "Barrier LOOP: Iteration %d BEGIN\n", iface->loop_state.counter);
-      lp_build_print_value(gallivm, "LOOP: InvocationId: \n", bld->system_values.invocation_id);
+   IRB()->SetInsertPoint(unwrap(LLVMGetInsertBlock(gallivm->builder)));
+
+   bld->system_values.invocation_id = wrap((LOAD(unwrap(iface->loop_var))));
+
+   if (verbose_tcs_shader_loop) {
+      lp_build_print_value(gallivm, "Barrier LOOP: Iteration BEGIN: ", iface->loop_var);
+      lp_build_print_value(gallivm, "Barrier LOOP: InvocationId: \n", bld->system_values.invocation_id);
    }
 }
 
@@ -2048,15 +2114,7 @@ BuilderSWR::CompileTCS(struct swr_context *ctx, swr_jit_tcs_key &key)
    system_values.prim_id =
       wrap(LOAD(pTcsCtx, {0, SWR_HS_CONTEXT_PrimitiveID}));
 
-   Constant *vInvocationId;
-   if (mVWidth == 8) {
-      vInvocationId = C({0, 1, 2, 3, 4, 5, 6, 7});
-   } else {
-      vInvocationId =
-         C({0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15});
-   }
-
-   system_values.invocation_id = wrap(vInvocationId);
+   system_values.invocation_id = wrap(VBROADCAST(C(0)));
    system_values.vertices_in = wrap(C(tcs->vertices_per_patch));
 
    if (verbose_shader) {
@@ -2243,7 +2301,7 @@ BuilderSWR::WriteVS(Value *pVal, Value *pVsContext, Value *pVtxOutput, unsigned
 #else
    Value *pOut = GEP(pVtxOutput, {0, 0, slot});
    STORE(pVal, pOut, {0, channel});
-   if (verbose_shader) {
+   if (verbose_vs_shader) {
       lp_build_printf(gallivm, "VS: Storing on slot %d, channel %d: ", C(slot), C(channel));
       lp_build_print_value(gallivm, "", wrap(pVal));
    }