Mesa (master): nv30: Support negative offsets in indirect constant access.

[Christoph Bumiller]

Module: Mesa
Branch: master
Commit: 542bd6941f5a56f7a3aa84b44d92591488b146bf
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=542bd6941f5a56f7a3aa84b44d92591488b146bf

Author: Roy Spliet <r.spliet at student.tudelft.nl>
Date:   Thu Jul 19 01:56:35 2012 +0200

nv30: Support negative offsets in indirect constant access.

Fixes piglit vp-address-01 amongst several others.

Signed-off-by: Roy Spliet <r.spliet at student.tudelft.nl>
Reviewed-by: Lucas Stach <dev at lynxeye.de>
Tested-by: Lucas Stach <dev at lynxeye.de>

---

 src/gallium/drivers/nv30/nv30_state.h    |    2 +-
 src/gallium/drivers/nv30/nv30_vertprog.c |    6 +++---
 src/gallium/drivers/nv30/nvfx_shader.h   |    2 +-
 src/gallium/drivers/nv30/nvfx_vertprog.c |   13 ++++++++++---
 4 files changed, 15 insertions(+), 8 deletions(-)

diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h
index 964676a..e27e16f 100644
--- a/src/gallium/drivers/nv30/nv30_state.h
+++ b/src/gallium/drivers/nv30/nv30_state.h
@@ -63,7 +63,7 @@ struct nv30_sampler_view {
 
 struct nv30_shader_reloc {
    unsigned location;
-   unsigned target;
+   int target;
 };
 
 struct nv30_vertprog_exec {
diff --git a/src/gallium/drivers/nv30/nv30_vertprog.c b/src/gallium/drivers/nv30/nv30_vertprog.c
index 9b5ba35..06e1b8c 100644
--- a/src/gallium/drivers/nv30/nv30_vertprog.c
+++ b/src/gallium/drivers/nv30/nv30_vertprog.c
@@ -149,8 +149,8 @@ nv30_vertprog_validate(struct nv30_context *nv30)
             inst     = vp->insns[reloc->location].data;
             target   = vp->data->start + reloc->target;
 
-            inst[1] &= ~0x0003fc000;
-            inst[1] |= target << 14;
+            inst[1] &= ~0x0007fc000;
+            inst[1] |= (target & 0x1ff) << 14;
             reloc++;
          }
       } else {
@@ -159,7 +159,7 @@ nv30_vertprog_validate(struct nv30_context *nv30)
             target   = vp->data->start + reloc->target;
 
             inst[1] &= ~0x0001ff000;
-            inst[1] |= target << 12;
+            inst[1] |= (target & 0x1ff) << 12;
             reloc++;
          }
       }
diff --git a/src/gallium/drivers/nv30/nvfx_shader.h b/src/gallium/drivers/nv30/nvfx_shader.h
index e343bf0..987e1b0 100644
--- a/src/gallium/drivers/nv30/nvfx_shader.h
+++ b/src/gallium/drivers/nv30/nvfx_shader.h
@@ -416,7 +416,7 @@
 
 struct nvfx_reg {
 	int8_t type;
-	uint32_t index;
+	int32_t index;
 };
 
 struct nvfx_src {
diff --git a/src/gallium/drivers/nv30/nvfx_vertprog.c b/src/gallium/drivers/nv30/nvfx_vertprog.c
index f41f82d..827d518 100644
--- a/src/gallium/drivers/nv30/nvfx_vertprog.c
+++ b/src/gallium/drivers/nv30/nvfx_vertprog.c
@@ -135,12 +135,13 @@ emit_src(struct nv30_context *nv30, struct nvfx_vpc *vpc, uint32_t *hw,
    case NVFXSR_CONST:
       sr |= (NVFX_VP(SRC_REG_TYPE_CONST) <<
              NVFX_VP(SRC_REG_TYPE_SHIFT));
-      if (src.reg.index < 512) {
+      if (src.reg.index < 256 && src.reg.index >= -256) {
          reloc.location = vp->nr_insns - 1;
          reloc.target = src.reg.index;
          util_dynarray_append(&vp->const_relocs, struct nvfx_relocation, reloc);
       } else {
-         hw[1] |= (src.reg.index - 512) << NVFX_VP(INST_CONST_SRC_SHIFT);
+         hw[1] |= (src.reg.index << NVFX_VP(INST_CONST_SRC_SHIFT)) &
+               NVFX_VP(INST_CONST_SRC_MASK);
       }
       break;
    case NVFXSR_NONE:
@@ -169,6 +170,7 @@ emit_src(struct nv30_context *nv30, struct nvfx_vpc *vpc, uint32_t *hw,
          hw[0] |= NVFX_VP(INST_INDEX_INPUT);
       else
          assert(0);
+
       if(src.indirect_reg)
          hw[0] |= NVFX_VP(INST_ADDR_REG_SELECT_1);
       hw[0] |= src.indirect_swz << NVFX_VP(INST_ADDR_SWZ_SHIFT);
@@ -367,7 +369,12 @@ tgsi_src(struct nvfx_vpc *vpc, const struct tgsi_full_src_register *fsrc) {
       src.reg = nvfx_reg(NVFXSR_INPUT, fsrc->Register.Index);
       break;
    case TGSI_FILE_CONSTANT:
-      src.reg = vpc->r_const[fsrc->Register.Index];
+      if(fsrc->Register.Indirect) {
+         src.reg = vpc->r_const[0];
+         src.reg.index = fsrc->Register.Index;
+      } else {
+         src.reg = vpc->r_const[fsrc->Register.Index];
+      }
       break;
    case TGSI_FILE_IMMEDIATE:
       src.reg = vpc->imm[fsrc->Register.Index];