#include "util/u_memory.h"
#include "util/u_simple_list.h"
#include "os/os_time.h"
+#include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_debug.h"
#include "gallivm/lp_bld_init.h"
#include "gallivm/lp_bld_intr.h"
#include "gallivm/lp_bld_flow.h"
+#include "gallivm/lp_bld_type.h"
#include <llvm-c/Analysis.h> /* for LLVMVerifyFunction */
#include "lp_perf.h"
#include "lp_flush.h"
#include "lp_screen.h"
#include "lp_context.h"
-#include "lp_setup_context.h"
-#include "lp_rast.h"
#include "lp_state.h"
#include "lp_state_fs.h"
#include "lp_state_setup.h"
LLVMValueRef dx20_ooa;
LLVMValueRef dx01_ooa;
- /* For twoside calcs
+ /* Temporary, per-attribute:
*/
- LLVMValueRef det;
- LLVMValueRef sign;
- LLVMValueRef bcolor_slot;
- LLVMValueRef color_slot;
-
+ LLVMValueRef v0a;
+ LLVMValueRef v1a;
+ LLVMValueRef v2a;
};
-static LLVMTypeRef type4f(void)
+
+
+static LLVMTypeRef
+type4f(struct gallivm_state *gallivm)
{
- return LLVMVectorType(LLVMFloatType(), 4);
+ return LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
}
/* Equivalent of _mm_setr_ps(a,b,c,d)
*/
-static LLVMValueRef vec4f(LLVMBuilderRef bld,
- LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
- const char *name)
+static LLVMValueRef
+vec4f(struct gallivm_state *gallivm,
+ LLVMValueRef a, LLVMValueRef b, LLVMValueRef c, LLVMValueRef d,
+ const char *name)
{
- LLVMValueRef i0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
- LLVMValueRef i1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
- LLVMValueRef i2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
- LLVMValueRef i3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+ LLVMBuilderRef bld = gallivm->builder;
+ LLVMValueRef i0 = lp_build_const_int32(gallivm, 0);
+ LLVMValueRef i1 = lp_build_const_int32(gallivm, 1);
+ LLVMValueRef i2 = lp_build_const_int32(gallivm, 2);
+ LLVMValueRef i3 = lp_build_const_int32(gallivm, 3);
- LLVMValueRef res = LLVMGetUndef(type4f());
+ LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
res = LLVMBuildInsertElement(bld, res, a, i0, "");
res = LLVMBuildInsertElement(bld, res, b, i1, "");
/* Equivalent of _mm_set1_ps(a)
*/
-static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
- LLVMValueRef a,
- const char *name)
+static LLVMValueRef
+vec4f_from_scalar(struct gallivm_state *gallivm,
+ LLVMValueRef a,
+ const char *name)
{
- LLVMValueRef res = LLVMGetUndef(type4f());
+ LLVMBuilderRef bld = gallivm->builder;
+ LLVMValueRef res = LLVMGetUndef(type4f(gallivm));
int i;
for(i = 0; i < 4; ++i) {
- LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+ LLVMValueRef index = lp_build_const_int32(gallivm, i);
res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
}
}
static void
-store_coef(LLVMBuilderRef builder,
+store_coef(struct gallivm_state *gallivm,
struct lp_setup_args *args,
unsigned slot,
LLVMValueRef a0,
LLVMValueRef dadx,
LLVMValueRef dady)
{
- LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), slot, 0);
+ LLVMBuilderRef builder = gallivm->builder;
+ LLVMValueRef idx = lp_build_const_int32(gallivm, slot);
LLVMBuildStore(builder,
a0,
static void
-emit_constant_coef4( LLVMBuilderRef builder,
+emit_constant_coef4(struct gallivm_state *gallivm,
struct lp_setup_args *args,
unsigned slot,
- LLVMValueRef vert,
- unsigned attr)
+ LLVMValueRef vert)
{
- LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
- LLVMValueRef zerovec = vec4f_from_scalar(builder, zero, "zero");
- LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), attr, 0);
- LLVMValueRef attr_ptr = LLVMBuildGEP(builder, vert, &idx, 1, "attr_ptr");
- LLVMValueRef vert_attr = LLVMBuildLoad(builder, attr_ptr, "vert_attr");
-
- store_coef(builder, args, slot, vert_attr, zerovec, zerovec);
+ LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+ LLVMValueRef zerovec = vec4f_from_scalar(gallivm, zero, "zero");
+ store_coef(gallivm, args, slot, vert, zerovec, zerovec);
}
* \param frontface is the triangle front facing?
*/
static void
-emit_facing_coef( LLVMBuilderRef builder,
+emit_facing_coef(struct gallivm_state *gallivm,
struct lp_setup_args *args,
unsigned slot )
{
+ LLVMBuilderRef builder = gallivm->builder;
+ LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
LLVMValueRef a0_0 = args->facing;
- LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, LLVMFloatType(), "");
- LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
- LLVMValueRef a0 = vec4f(builder, a0_0f, zero, zero, zero, "facing");
- LLVMValueRef zerovec = vec4f_from_scalar(builder, zero, "zero");
+ LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, float_type, "");
+ LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+ LLVMValueRef a0 = vec4f(gallivm, a0_0f, zero, zero, zero, "facing");
+ LLVMValueRef zerovec = vec4f_from_scalar(gallivm, zero, "zero");
- store_coef(builder, args, slot, a0, zerovec, zerovec);
+ store_coef(gallivm, args, slot, a0, zerovec, zerovec);
}
static LLVMValueRef
-vert_attrib(LLVMBuilderRef b,
+vert_attrib(struct gallivm_state *gallivm,
LLVMValueRef vert,
int attr,
int elem,
const char *name)
{
+ LLVMBuilderRef b = gallivm->builder;
LLVMValueRef idx[2];
- idx[0] = LLVMConstInt(LLVMInt32Type(), attr, 0);
- idx[1] = LLVMConstInt(LLVMInt32Type(), elem, 0);
+ idx[0] = lp_build_const_int32(gallivm, attr);
+ idx[1] = lp_build_const_int32(gallivm, elem);
return LLVMBuildLoad(b, LLVMBuildGEP(b, vert, idx, 2, ""), name);
}
+static LLVMValueRef
+vert_clamp(LLVMBuilderRef b,
+ LLVMValueRef x,
+ LLVMValueRef min,
+ LLVMValueRef max)
+{
+ LLVMValueRef min_result = LLVMBuildFCmp(b, LLVMRealUGT, min, x, "");
+ LLVMValueRef max_result = LLVMBuildFCmp(b, LLVMRealUGT, x, max, "");
+ LLVMValueRef clamp_value;
+
+ clamp_value = LLVMBuildSelect(b, min_result, min, x, "");
+ clamp_value = LLVMBuildSelect(b, max_result, max, x, "");
+ return clamp_value;
+}
+
+static void
+lp_twoside(struct gallivm_state *gallivm,
+ struct lp_setup_args *args,
+ const struct lp_setup_variant_key *key,
+ int bcolor_slot)
+{
+ LLVMBuilderRef b = gallivm->builder;
+ LLVMValueRef a0_back, a1_back, a2_back;
+ LLVMValueRef idx2 = lp_build_const_int32(gallivm, bcolor_slot);
+
+ LLVMValueRef facing = args->facing;
+ LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, lp_build_const_int32(gallivm, 0), ""); /** need i1 for if condition */
+
+ a0_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a_back");
+ a1_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a_back");
+ a2_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx2, 1, ""), "v2a_back");
+
+ /* Possibly swap the front and back attrib values,
+ *
+ * Prefer select to if so we don't have to worry about phis or
+ * allocas.
+ */
+ args->v0a = LLVMBuildSelect(b, front_facing, a0_back, args->v0a, "");
+ args->v1a = LLVMBuildSelect(b, front_facing, a1_back, args->v1a, "");
+ args->v2a = LLVMBuildSelect(b, front_facing, a2_back, args->v2a, "");
+
+}
+
+static void
+lp_do_offset_tri(struct gallivm_state *gallivm,
+ struct lp_setup_args *args,
+ const struct lp_setup_variant_key *key)
+{
+ LLVMBuilderRef b = gallivm->builder;
+ struct lp_build_context bld;
+ LLVMValueRef zoffset, mult;
+ LLVMValueRef z0_new, z1_new, z2_new;
+ LLVMValueRef dzdx0, dzdx, dzdy0, dzdy;
+ LLVMValueRef max, max_value;
+
+ LLVMValueRef one = lp_build_const_float(gallivm, 1.0);
+ LLVMValueRef zero = lp_build_const_float(gallivm, 0.0);
+ LLVMValueRef two = lp_build_const_int32(gallivm, 2);
+
+ /* edge vectors: e = v0 - v2, f = v1 - v2 */
+ LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+ LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+ LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+ LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
+ LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
+ LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
+ LLVMValueRef v0_z = vert_attrib(gallivm, args->v0, 0, 2, "v0_z");
+ LLVMValueRef v1_z = vert_attrib(gallivm, args->v1, 0, 2, "v1_z");
+ LLVMValueRef v2_z = vert_attrib(gallivm, args->v2, 0, 2, "v2_z");
+
+ /* edge vectors: e = v0 - v2, f = v1 - v2 */
+ LLVMValueRef dx02 = LLVMBuildFSub(b, v0_x, v2_x, "dx02");
+ LLVMValueRef dy02 = LLVMBuildFSub(b, v0_y, v2_y, "dy02");
+ LLVMValueRef dz02 = LLVMBuildFSub(b, v0_z, v2_z, "dz02");
+ LLVMValueRef dx12 = LLVMBuildFSub(b, v1_x, v2_x, "dx12");
+ LLVMValueRef dy12 = LLVMBuildFSub(b, v1_y, v2_y, "dy12");
+ LLVMValueRef dz12 = LLVMBuildFSub(b, v1_z, v2_z, "dz12");
+
+ /* det = cross(e,f).z */
+ LLVMValueRef dx02_dy12 = LLVMBuildFMul(b, dx02, dy12, "dx02_dy12");
+ LLVMValueRef dy02_dx12 = LLVMBuildFMul(b, dy02, dx12, "dy02_dx12");
+ LLVMValueRef det = LLVMBuildFSub(b, dx02_dy12, dy02_dx12, "det");
+ LLVMValueRef inv_det = LLVMBuildFDiv(b, one, det, "inv_det");
+
+ /* (res1,res2) = cross(e,f).xy */
+ LLVMValueRef dy02_dz12 = LLVMBuildFMul(b, dy02, dz12, "dy02_dz12");
+ LLVMValueRef dz02_dy12 = LLVMBuildFMul(b, dz02, dy12, "dz02_dy12");
+ LLVMValueRef dz02_dx12 = LLVMBuildFMul(b, dz02, dx12, "dz02_dx12");
+ LLVMValueRef dx02_dz12 = LLVMBuildFMul(b, dx02, dz12, "dx02_dz12");
+ LLVMValueRef res1 = LLVMBuildFSub(b, dy02_dz12, dz02_dy12, "res1");
+ LLVMValueRef res2 = LLVMBuildFSub(b, dz02_dx12, dx02_dz12, "res2");
+
+ /* dzdx = fabsf(res1 * inv_det), dydx = fabsf(res2 * inv_det)*/
+ lp_build_context_init(&bld, gallivm, lp_type_float(32));
+ dzdx0 = LLVMBuildFMul(b, res1, inv_det, "dzdx");
+ dzdx = lp_build_abs(&bld, dzdx0);
+ dzdy0 = LLVMBuildFMul(b, res2, inv_det, "dzdy");
+ dzdy = lp_build_abs(&bld, dzdy0);
+
+ /* zoffset = offset->units + MAX2(dzdx, dzdy) * offset->scale */
+ max = LLVMBuildFCmp(b, LLVMRealUGT, dzdx, dzdy, "");
+ max_value = LLVMBuildSelect(b, max, dzdx, dzdy, "max");
+
+ mult = LLVMBuildFMul(b, max_value, lp_build_const_float(gallivm, key->scale), "");
+ zoffset = LLVMBuildFAdd(b, lp_build_const_float(gallivm, key->units), mult, "zoffset");
+
+ /* clamp and do offset */
+ z0_new = vert_clamp(b, LLVMBuildFAdd(b, v0_z, zoffset, ""), zero, one);
+ z1_new = vert_clamp(b, LLVMBuildFAdd(b, v1_z, zoffset, ""), zero, one);
+ z2_new = vert_clamp(b, LLVMBuildFAdd(b, v2_z, zoffset, ""), zero, one);
+
+ /* insert into args->a0.z, a1.z, a2.z:
+ */
+ args->v0a = LLVMBuildInsertElement(b, args->v0a, z0_new, two, "");
+ args->v1a = LLVMBuildInsertElement(b, args->v1a, z1_new, two, "");
+ args->v2a = LLVMBuildInsertElement(b, args->v2a, z2_new, two, "");
+}
+
+static void
+load_attribute(struct gallivm_state *gallivm,
+ struct lp_setup_args *args,
+ const struct lp_setup_variant_key *key,
+ unsigned vert_attr)
+{
+ LLVMBuilderRef b = gallivm->builder;
+ LLVMValueRef idx = lp_build_const_int32(gallivm, vert_attr);
+
+ /* Load the vertex data
+ */
+ args->v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
+ args->v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
+ args->v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
+
+
+ /* Potentially modify it according to twoside, offset, etc:
+ */
+ if (vert_attr == 0 && (key->scale != 0.0f || key->units != 0.0f)) {
+ lp_do_offset_tri(gallivm, args, key);
+ }
+
+ if (key->twoside) {
+ if (vert_attr == key->color_slot && key->bcolor_slot >= 0)
+ lp_twoside(gallivm, args, key, key->bcolor_slot);
+ else if (vert_attr == key->spec_slot && key->bspec_slot >= 0)
+ lp_twoside(gallivm, args, key, key->bspec_slot);
+ }
+}
static void
-emit_coef4( LLVMBuilderRef b,
+emit_coef4( struct gallivm_state *gallivm,
struct lp_setup_args *args,
unsigned slot,
LLVMValueRef a0,
LLVMValueRef a1,
LLVMValueRef a2)
{
+ LLVMBuilderRef b = gallivm->builder;
LLVMValueRef dy20_ooa = args->dy20_ooa;
LLVMValueRef dy01_ooa = args->dy01_ooa;
LLVMValueRef dx20_ooa = args->dx20_ooa;
LLVMValueRef attr_v0 = LLVMBuildFAdd(b, dadx_x0, dady_y0, "attr_v0");
LLVMValueRef attr_0 = LLVMBuildFSub(b, a0, attr_v0, "attr_0");
- store_coef(b, args, slot, attr_0, dadx, dady);
+ store_coef(gallivm, args, slot, attr_0, dadx, dady);
}
static void
-emit_linear_coef( LLVMBuilderRef b,
+emit_linear_coef( struct gallivm_state *gallivm,
struct lp_setup_args *args,
- unsigned slot,
- unsigned vert_attr)
+ unsigned slot)
{
- LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
-
- LLVMValueRef a0 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
- LLVMValueRef a1 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
- LLVMValueRef a2 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
-
- emit_coef4(b, args, slot, a0, a1, a2);
+ /* nothing to do anymore */
+ emit_coef4(gallivm,
+ args, slot,
+ args->v0a,
+ args->v1a,
+ args->v2a);
}
-
/**
* Compute a0, dadx and dady for a perspective-corrected interpolant,
* for a triangle.
* divide the interpolated value by the interpolated W at that fragment.
*/
static void
-emit_perspective_coef( LLVMBuilderRef b,
+emit_perspective_coef( struct gallivm_state *gallivm,
struct lp_setup_args *args,
- unsigned slot,
- unsigned vert_attr)
+ unsigned slot)
{
+ LLVMBuilderRef b = gallivm->builder;
+
/* premultiply by 1/w (v[0][3] is always 1/w):
*/
- LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
+ LLVMValueRef v0_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v0, 0, 3, ""), "v0_oow");
+ LLVMValueRef v1_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v1, 0, 3, ""), "v1_oow");
+ LLVMValueRef v2_oow = vec4f_from_scalar(gallivm, vert_attrib(gallivm, args->v2, 0, 3, ""), "v2_oow");
- LLVMValueRef v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
- LLVMValueRef v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
- LLVMValueRef v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
+ LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, args->v0a, v0_oow, "v0_oow_v0a");
+ LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, args->v1a, v1_oow, "v1_oow_v1a");
+ LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, args->v2a, v2_oow, "v2_oow_v2a");
- LLVMValueRef v0_oow = vec4f_from_scalar(b, vert_attrib(b, args->v0, 0, 3, ""), "v0_oow");
- LLVMValueRef v1_oow = vec4f_from_scalar(b, vert_attrib(b, args->v1, 0, 3, ""), "v1_oow");
- LLVMValueRef v2_oow = vec4f_from_scalar(b, vert_attrib(b, args->v2, 0, 3, ""), "v2_oow");
-
- LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, v0a, v0_oow, "v0_oow_v0a");
- LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, v1a, v1_oow, "v1_oow_v1a");
- LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, v2a, v2_oow, "v2_oow_v2a");
-
- emit_coef4(b, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
+ emit_coef4(gallivm, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
}
static void
-emit_position_coef( LLVMBuilderRef builder,
+emit_position_coef( struct gallivm_state *gallivm,
struct lp_setup_args *args,
- int slot, int attrib )
+ int slot )
{
- emit_linear_coef(builder, args, slot, attrib);
+ emit_linear_coef(gallivm, args, slot);
}
* Compute the inputs-> dadx, dady, a0 values.
*/
static void
-emit_tri_coef( LLVMBuilderRef builder,
+emit_tri_coef( struct gallivm_state *gallivm,
const struct lp_setup_variant_key *key,
struct lp_setup_args *args )
{
/* The internal position input is in slot zero:
*/
- emit_position_coef(builder, args, 0, 0);
+ load_attribute(gallivm, args, key, 0);
+ emit_position_coef(gallivm, args, 0);
/* setup interpolation for all the remaining attributes:
*/
for (slot = 0; slot < key->num_inputs; slot++) {
- unsigned vert_attr = key->inputs[slot].src_index;
+
+ if (key->inputs[slot].interp == LP_INTERP_CONSTANT ||
+ key->inputs[slot].interp == LP_INTERP_LINEAR ||
+ key->inputs[slot].interp == LP_INTERP_PERSPECTIVE)
+ load_attribute(gallivm, args, key, key->inputs[slot].src_index);
switch (key->inputs[slot].interp) {
case LP_INTERP_CONSTANT:
if (key->flatshade_first) {
- emit_constant_coef4(builder, args, slot+1, args->v0, vert_attr);
+ emit_constant_coef4(gallivm, args, slot+1, args->v0a);
}
else {
- emit_constant_coef4(builder, args, slot+1, args->v2, vert_attr);
+ emit_constant_coef4(gallivm, args, slot+1, args->v2a);
}
break;
case LP_INTERP_LINEAR:
- emit_linear_coef(builder, args, slot+1, vert_attr);
+ emit_linear_coef(gallivm, args, slot+1);
break;
case LP_INTERP_PERSPECTIVE:
- emit_perspective_coef(builder, args, slot+1, vert_attr);
+ emit_perspective_coef(gallivm, args, slot+1);
break;
case LP_INTERP_POSITION:
break;
case LP_INTERP_FACING:
- emit_facing_coef(builder, args, slot+1);
+ emit_facing_coef(gallivm, args, slot+1);
break;
default:
/* XXX: This is generic code, share with fs/vs codegen:
*/
static lp_jit_setup_triangle
-finalize_function(struct llvmpipe_screen *screen,
+finalize_function(struct gallivm_state *gallivm,
LLVMBuilderRef builder,
LLVMValueRef function)
{
#endif
/* Apply optimizations to LLVM IR */
- LLVMRunFunctionPassManager(screen->pass, function);
+ LLVMRunFunctionPassManager(gallivm->passmgr, function);
if (gallivm_debug & GALLIVM_DEBUG_IR)
{
/*
* Translate the LLVM IR into machine code.
*/
- f = LLVMGetPointerToGlobal(screen->engine, function);
+ f = LLVMGetPointerToGlobal(gallivm->engine, function);
if (gallivm_debug & GALLIVM_DEBUG_ASM)
{
lp_func_delete_body(function);
- return f;
+ return (lp_jit_setup_triangle) pointer_to_func(f);
}
/* XXX: Generic code:
*/
static void
-lp_emit_emms(LLVMBuilderRef builder)
+lp_emit_emms(struct gallivm_state *gallivm)
{
#ifdef PIPE_ARCH_X86
/* Avoid corrupting the FPU stack on 32bit OSes. */
- lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
+ lp_build_intrinsic(gallivm->builder, "llvm.x86.mmx.emms",
+ LLVMVoidTypeInContext(gallivm->context), NULL, 0);
#endif
}
}
static void
-init_args(LLVMBuilderRef b,
+init_args(struct gallivm_state *gallivm,
struct lp_setup_args *args,
const struct lp_setup_variant *variant)
{
- LLVMValueRef v0_x = vert_attrib(b, args->v0, 0, 0, "v0_x");
- LLVMValueRef v0_y = vert_attrib(b, args->v0, 0, 1, "v0_y");
+ LLVMBuilderRef b = gallivm->builder;
- LLVMValueRef v1_x = vert_attrib(b, args->v1, 0, 0, "v1_x");
- LLVMValueRef v1_y = vert_attrib(b, args->v1, 0, 1, "v1_y");
+ LLVMValueRef v0_x = vert_attrib(gallivm, args->v0, 0, 0, "v0_x");
+ LLVMValueRef v0_y = vert_attrib(gallivm, args->v0, 0, 1, "v0_y");
- LLVMValueRef v2_x = vert_attrib(b, args->v2, 0, 0, "v2_x");
- LLVMValueRef v2_y = vert_attrib(b, args->v2, 0, 1, "v2_y");
+ LLVMValueRef v1_x = vert_attrib(gallivm, args->v1, 0, 0, "v1_x");
+ LLVMValueRef v1_y = vert_attrib(gallivm, args->v1, 0, 1, "v1_y");
- LLVMValueRef pixel_center = LLVMConstReal(LLVMFloatType(),
- variant->key.pixel_center_half ? 0.5 : 0);
+ LLVMValueRef v2_x = vert_attrib(gallivm, args->v2, 0, 0, "v2_x");
+ LLVMValueRef v2_y = vert_attrib(gallivm, args->v2, 0, 1, "v2_y");
+
+ LLVMValueRef pixel_center = lp_build_const_float(gallivm,
+ variant->key.pixel_center_half ? 0.5 : 0);
LLVMValueRef x0_center = LLVMBuildFSub(b, v0_x, pixel_center, "x0_center" );
LLVMValueRef y0_center = LLVMBuildFSub(b, v0_y, pixel_center, "y0_center" );
LLVMValueRef dx20 = LLVMBuildFSub(b, v2_x, v0_x, "dx20");
LLVMValueRef dy20 = LLVMBuildFSub(b, v2_y, v0_y, "dy20");
- LLVMValueRef one = LLVMConstReal(LLVMFloatType(), 1.0);
+ LLVMValueRef one = lp_build_const_float(gallivm, 1.0);
LLVMValueRef e = LLVMBuildFMul(b, dx01, dy20, "e");
LLVMValueRef f = LLVMBuildFMul(b, dx20, dy01, "f");
LLVMValueRef ooa = LLVMBuildFDiv(b, one, LLVMBuildFSub(b, e, f, ""), "ooa");
LLVMValueRef dx20_ooa = LLVMBuildFMul(b, dx20, ooa, "dx20_ooa");
LLVMValueRef dx01_ooa = LLVMBuildFMul(b, dx01, ooa, "dx01_ooa");
- args->dy20_ooa = vec4f_from_scalar(b, dy20_ooa, "dy20_ooa_4f");
- args->dy01_ooa = vec4f_from_scalar(b, dy01_ooa, "dy01_ooa_4f");
-
- args->dx20_ooa = vec4f_from_scalar(b, dx20_ooa, "dx20_ooa_4f");
- args->dx01_ooa = vec4f_from_scalar(b, dx01_ooa, "dx01_ooa_4f");
-
- args->x0_center = vec4f_from_scalar(b, x0_center, "x0_center_4f");
- args->y0_center = vec4f_from_scalar(b, y0_center, "y0_center_4f");
-}
-
-static void
-set_args_attr(struct llvmpipe_context *lp,
- struct lp_setup_args *args)
-{
- args->color_slot = LLVMConstInt(LLVMInt32Type(), lp->color_slot, 0);
- args->bcolor_slot = LLVMConstInt(LLVMInt32Type(), lp->bcolor_slot, 0);
- args->sign = LLVMConstReal(LLVMFloatType(), (lp->rasterizer->front_ccw ? -1.0f : 1.0f));
-}
-
-static void
-lp_twoside(LLVMBuilderRef b,
- struct lp_setup_args *args,
- const struct lp_setup_variant_key *key)
-{
- struct lp_build_if_state if_state;
-
- LLVMValueRef a0_old, a1_old, a2_old;
- LLVMValueRef a0_new, a1_new, a2_new;
+ args->dy20_ooa = vec4f_from_scalar(gallivm, dy20_ooa, "dy20_ooa_4f");
+ args->dy01_ooa = vec4f_from_scalar(gallivm, dy01_ooa, "dy01_ooa_4f");
- LLVMValueRef idx1 = args->color_slot;
- LLVMValueRef idx2 = args->bcolor_slot;
+ args->dx20_ooa = vec4f_from_scalar(gallivm, dx20_ooa, "dx20_ooa_4f");
+ args->dx01_ooa = vec4f_from_scalar(gallivm, dx01_ooa, "dx01_ooa_4f");
- LLVMValueRef facing = args->facing;
- LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, LLVMConstInt(LLVMInt32Type(), 0, 0), ""); /** need i1 for loop condition */
-
-#if 0
-/*Probably can delete this, just tried to follow draw_pipe_twoside way of
- calculating det*/
- /* edge vectors: e = v0 - v2, f = v1 - v2 */
- LLVMValueRef e = LLVMBuildFSub(b, args->v0, args->v2, "e");
- LLVMValueRef f = LLVMBuildFSub(b, args->v1, args->v2, "f");
- LLVMValueRef dx02 = vert_attrib(b, e, 0, 0, "dx02");
- LLVMValueRef dy02 = vert_attrib(b, e, 0, 1, "dy02");
- LLVMValueRef dx12 = vert_attrib(b, f, 0, 0, "dx12");
- LLVMValueRef dy12 = vert_attrib(b, f, 0, 1, "dy12");
-
- /* det = cross(e,f).z */
- LLVMValueRef dx02_dy12 = LLVMBuildFMul(b, dx02, dy12, "dx02_dy12");
- LLVMValueRef dy02_dx12 = LLVMBuildFMul(b, dy02, dx12, "dy02_dx12");
- LLVMValueRef det = LLVMBuildFSub(b, dx02_dy12, dy02_dx12, "det");
- args->det = det;
- LLVMValueRef result = LLVMBuildFMul(b, det, args->sign, "dy02_dx12");
-#endif
-
- lp_build_if(&if_state, b, front_facing);
- {
- /* swap the front and back attrib values */
- a0_old = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx1, 1, ""), "v0a");
- a1_old = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx1, 1, ""), "v1a");
- a2_old = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx1, 1, ""), "v2a");
-
- a0_new = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a");
- a1_new = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a");
- a2_new = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx2, 1, ""), "v2a");
-
- LLVMBuildStore(b, a0_new, LLVMBuildGEP(b, args->v0, &idx1, 1, ""));
- LLVMBuildStore(b, a1_new, LLVMBuildGEP(b, args->v1, &idx1, 1, ""));
- LLVMBuildStore(b, a2_new, LLVMBuildGEP(b, args->v2, &idx1, 1, ""));
- }
- lp_build_endif(&if_state);
-
+ args->x0_center = vec4f_from_scalar(gallivm, x0_center, "x0_center_4f");
+ args->y0_center = vec4f_from_scalar(gallivm, y0_center, "y0_center_4f");
}
-
/**
* Generate the runtime callable function for the coefficient calculation.
*
*/
static struct lp_setup_variant *
-generate_setup_variant(struct llvmpipe_screen *screen,
+generate_setup_variant(struct gallivm_state *gallivm,
struct lp_setup_variant_key *key,
struct llvmpipe_context *lp)
{
LLVMTypeRef func_type;
LLVMTypeRef arg_types[7];
LLVMBasicBlockRef block;
- LLVMBuilderRef builder;
- int64_t t0, t1;
+ LLVMBuilderRef builder = gallivm->builder;
+ int64_t t0 = 0, t1;
if (0)
goto fail;
* the vertices.
*/
- vec4f_type = LLVMVectorType(LLVMFloatType(), 4);
+ vec4f_type = LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);
arg_types[0] = LLVMPointerType(vec4f_type, 0); /* v0 */
arg_types[1] = LLVMPointerType(vec4f_type, 0); /* v1 */
arg_types[2] = LLVMPointerType(vec4f_type, 0); /* v2 */
- arg_types[3] = LLVMInt32Type(); /* facing */
+ arg_types[3] = LLVMInt32TypeInContext(gallivm->context); /* facing */
arg_types[4] = LLVMPointerType(vec4f_type, 0); /* a0, aligned */
arg_types[5] = LLVMPointerType(vec4f_type, 0); /* dadx, aligned */
arg_types[6] = LLVMPointerType(vec4f_type, 0); /* dady, aligned */
- func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+ func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
+ arg_types, Elements(arg_types), 0);
- variant->function = LLVMAddFunction(screen->module, func_name, func_type);
+ variant->function = LLVMAddFunction(gallivm->module, func_name, func_type);
if (!variant->function)
goto fail;
/*
* Function body
*/
- block = LLVMAppendBasicBlock(variant->function, "entry");
- builder = LLVMCreateBuilder();
+ block = LLVMAppendBasicBlockInContext(gallivm->context,
+ variant->function, "entry");
LLVMPositionBuilderAtEnd(builder, block);
set_noalias(builder, variant->function, arg_types, Elements(arg_types));
- init_args(builder, &args, variant);
- if (variant->key.twoside){
- set_args_attr(lp, &args);
- lp_twoside(builder, &args, &variant->key);
- }
- emit_tri_coef(builder, &variant->key, &args);
+ init_args(gallivm, &args, variant);
+ emit_tri_coef(gallivm, &variant->key, &args);
- lp_emit_emms(builder);
+ lp_emit_emms(gallivm);
LLVMBuildRetVoid(builder);
- LLVMDisposeBuilder(builder);
- variant->jit_function = finalize_function(screen, builder,
+ variant->jit_function = finalize_function(gallivm, builder,
variant->function);
if (!variant->jit_function)
goto fail;
if (variant) {
if (variant->function) {
if (variant->jit_function)
- LLVMFreeMachineCodeForFunction(screen->engine,
+ LLVMFreeMachineCodeForFunction(gallivm->engine,
variant->function);
LLVMDeleteFunction(variant->function);
}
key->twoside = lp->rasterizer->light_twoside;
key->size = Offset(struct lp_setup_variant_key,
inputs[key->num_inputs]);
- key->pad = 0;
+ key->color_slot = lp->color_slot [0];
+ key->bcolor_slot = lp->bcolor_slot[0];
+ key->spec_slot = lp->color_slot [1];
+ key->bspec_slot = lp->bcolor_slot[1];
+ assert(key->color_slot == lp->color_slot [0]);
+ assert(key->bcolor_slot == lp->bcolor_slot[0]);
+ assert(key->spec_slot == lp->color_slot [1]);
+ assert(key->bspec_slot == lp->bcolor_slot[1]);
+
+ key->units = (float) (lp->rasterizer->offset_units * lp->mrd);
+ key->scale = lp->rasterizer->offset_scale;
+ key->pad = 0;
memcpy(key->inputs, fs->inputs, key->num_inputs * sizeof key->inputs[0]);
for (i = 0; i < key->num_inputs; i++) {
if (key->inputs[i].interp == LP_INTERP_COLOR) {
remove_setup_variant(struct llvmpipe_context *lp,
struct lp_setup_variant *variant)
{
- struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
if (gallivm_debug & GALLIVM_DEBUG_IR) {
debug_printf("llvmpipe: del setup_variant #%u total %u\n",
variant->no, lp->nr_setup_variants);
if (variant->function) {
if (variant->jit_function)
- LLVMFreeMachineCodeForFunction(screen->engine,
+ LLVMFreeMachineCodeForFunction(lp->gallivm->engine,
variant->function);
LLVMDeleteFunction(variant->function);
}
llvmpipe_finish(pipe, __FUNCTION__);
for (i = 0; i < LP_MAX_SETUP_VARIANTS / 4; i++) {
- struct lp_setup_variant_list_item *item = last_elem(&lp->setup_variants_list);
+ struct lp_setup_variant_list_item *item;
+ if (is_empty_list(&lp->setup_variants_list)) {
+ break;
+ }
+ item = last_elem(&lp->setup_variants_list);
+ assert(item);
+ assert(item->base);
remove_setup_variant(lp, item->base);
}
}
void
llvmpipe_update_setup(struct llvmpipe_context *lp)
{
- struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
-
struct lp_setup_variant_key *key = &lp->setup_variant.key;
struct lp_setup_variant *variant = NULL;
struct lp_setup_variant_list_item *li;
cull_setup_variants(lp);
}
- variant = generate_setup_variant(screen, key, lp);
- insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
- lp->nr_setup_variants++;
+ variant = generate_setup_variant(lp->gallivm, key, lp);
+ if (variant) {
+ insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
+ lp->nr_setup_variants++;
+ llvmpipe_variant_count++;
+ }
}
lp_setup_set_setup_variant(lp->setup,
{
int i, slot;
- for (i = 0; i < NUM_CHANNELS; i++) {
+ for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
float a0 = sa0 [0][i];
float dadx = sdadx[0][i];
float dady = sdady[0][i];
for (slot = 0; slot < key->num_inputs; slot++) {
unsigned usage_mask = key->inputs[slot].usage_mask;
- for (i = 0; i < NUM_CHANNELS; i++) {
+ for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
if (usage_mask & (1 << i)) {
float a0 = sa0 [1 + slot][i];
float dadx = sdadx[1 + slot][i];
projects / mesa.git / blobdiff