gallivm: Pass texture coords derivates as scalars.

We end up treating them as scalars in the end, and it saves some
instructions.

4 files changed, 38 insertions, 26 deletions

diff --git a/src/gallium/auxiliary/gallivm/lp_bld_quad.c b/src/gallium/auxiliary/gallivm/lp_bld_quad.c
index 7b10889..c18c8b4 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_quad.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_quad.c
@@ -81,11 +81,15 @@ LLVMValueRef
 lp_build_scalar_ddx(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_left  = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_right = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_RIGHT, 0);
-   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "");
-   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "");
-   return lp_build_sub(bld, a_right, a_left);
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef idx_left  = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMValueRef idx_right = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_RIGHT, 0);
+   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "left");
+   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "right");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_right, a_left, "ddx");
+   else
+      return LLVMBuildSub(bld->builder, a_right, a_left, "ddx");
 }
 
 
@@ -93,9 +97,13 @@ LLVMValueRef
 lp_build_scalar_ddy(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_top    = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_bottom = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_BOTTOM_LEFT, 0);
-   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "");
-   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "");
-   return lp_build_sub(bld, a_bottom, a_top);
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef idx_top    = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMValueRef idx_bottom = LLVMConstInt(i32t, LP_BLD_QUAD_BOTTOM_LEFT, 0);
+   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "top");
+   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "bottom");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_bottom, a_top, "ddy");
+   else
+      return LLVMBuildSub(bld->builder, a_bottom, a_top, "ddy");
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
index 9d15a6f..844d1d9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
@@ -200,8 +200,8 @@ lp_build_rho(struct lp_build_sample_context *bld,
    LLVMValueRef float_size;
    LLVMValueRef rho;
 
-   dsdx = LLVMBuildExtractElement(bld->builder, ddx[0], index0, "dsdx");
-   dsdy = LLVMBuildExtractElement(bld->builder, ddy[0], index0, "dsdy");
+   dsdx = ddx[0];
+   dsdy = ddy[0];
 
    if (dims <= 1) {
       rho_x = dsdx;
@@ -214,15 +214,15 @@ lp_build_rho(struct lp_build_sample_context *bld,
       rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dsdx, index0, "");
       rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dsdy, index0, "");
 
-      dtdx = LLVMBuildExtractElement(bld->builder, ddx[1], index0, "dtdx");
-      dtdy = LLVMBuildExtractElement(bld->builder, ddy[1], index0, "dtdy");
+      dtdx = ddx[1];
+      dtdy = ddy[1];
 
       rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dtdx, index1, "");
       rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dtdy, index1, "");
 
       if (dims >= 3) {
-         drdx = LLVMBuildExtractElement(bld->builder, ddx[2], index0, "drdx");
-         drdy = LLVMBuildExtractElement(bld->builder, ddy[2], index0, "drdy");
+         drdx = ddx[2];
+         drdy = ddy[2];
 
          rho_x = LLVMBuildInsertElement(bld->builder, rho_x, drdx, index2, "");
          rho_y = LLVMBuildInsertElement(bld->builder, rho_y, drdy, index2, "");
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
index 69d81ad..be5d9a2 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
@@ -942,12 +942,12 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
       r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
 
       /* recompute ddx, ddy using the new (s,t) face texcoords */
-      face_ddx[0] = lp_build_ddx(&bld->coord_bld, s);
-      face_ddx[1] = lp_build_ddx(&bld->coord_bld, t);
+      face_ddx[0] = lp_build_scalar_ddx(&bld->coord_bld, s);
+      face_ddx[1] = lp_build_scalar_ddx(&bld->coord_bld, t);
       face_ddx[2] = NULL;
       face_ddx[3] = NULL;
-      face_ddy[0] = lp_build_ddy(&bld->coord_bld, s);
-      face_ddy[1] = lp_build_ddy(&bld->coord_bld, t);
+      face_ddy[0] = lp_build_scalar_ddy(&bld->coord_bld, s);
+      face_ddy[1] = lp_build_scalar_ddy(&bld->coord_bld, t);
       face_ddy[2] = NULL;
       face_ddy[3] = NULL;
       ddx = face_ddx;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 03020a6..2d4eb9a 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -887,21 +887,25 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
    }
 
    if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
+      LLVMTypeRef i32t = LLVMInt32Type();
+      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
       for (i = 0; i < num_coords; i++) {
-         ddx[i] = emit_fetch( bld, inst, 1, i );
-         ddy[i] = emit_fetch( bld, inst, 2, i );
+         LLVMValueRef src1 = emit_fetch( bld, inst, 1, i );
+         LLVMValueRef src2 = emit_fetch( bld, inst, 2, i );
+         ddx[i] = LLVMBuildExtractElement(bld->base.builder, src1, index0, "");
+         ddy[i] = LLVMBuildExtractElement(bld->base.builder, src2, index0, "");
       }
       unit = inst->Src[3].Register.Index;
    }  else {
       for (i = 0; i < num_coords; i++) {
-         ddx[i] = lp_build_ddx( &bld->base, coords[i] );
-         ddy[i] = lp_build_ddy( &bld->base, coords[i] );
+         ddx[i] = lp_build_scalar_ddx( &bld->base, coords[i] );
+         ddy[i] = lp_build_scalar_ddy( &bld->base, coords[i] );
       }
       unit = inst->Src[1].Register.Index;
    }
    for (i = num_coords; i < 3; i++) {
-      ddx[i] = bld->base.undef;
-      ddy[i] = bld->base.undef;
+      ddx[i] = LLVMGetUndef(bld->base.elem_type);
+      ddy[i] = LLVMGetUndef(bld->base.elem_type);
    }
 
    bld->sampler->emit_fetch_texel(bld->sampler,